Marginal Cost Water Pricing: Welfare Effects and Policy Implications

using Minimum Cost and Benchmarking Models,

with Case Studies from Australia and Asia

Thesis

Submitted to the School of Economics

of the University of Adelaide
in fulfilment of the requirements

for the degree of

Doctor of Philosophy

by

David Altmann, BA. MEngSci. GDip.Econ.

Dr. Eran Binenbaum, Principal Supervisor

Dr. John Hatch & Dr. Ralph Bayer, Co-supervisors

Adelaide, Australia
August 2007
for Marcella
 CONTENTS

FIGURES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viii

TABLES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix

ABBREVIATIONS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi

ABSTRACT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiv

THESIS DECLARATION . . . . . . . . . . . . . . . . . . . . . . . . . . . xvi

ACKNOWLEDGMENTS . . . . . . . . . . . . . . . . . . . . . . . . . . . xvii

CHAPTER 1: Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 1

CHAPTER 2: Urban Water Pricing . . . . . . . . . . . . . . . . . . . . . 8

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.2 Theoretical Foundations of Utility Pricing - First and Second Best
Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.3 The Regulatory Environment . . . . . . . . . . . . . . . . . . . . 11
2.3.1 Cost of Service Regulation . . . . . . . . . . . . . . . . . . 13
2.3.2 Price Cap Regulation . . . . . . . . . . . . . . . . . . . . . 15
2.3.3 Performance and Yardstick Based Approaches . . . . . . . 16
2.3.4 Incentive Regulation Theory . . . . . . . . . . . . . . . . . 16
2.4 Practical Approaches to Urban Water Pricing . . . . . . . . . . . 18
2.4.1 Turvey Long Run Marginal Cost . . . . . . . . . . . . . . 20
2.4.2 Average Incremental Cost Pricing . . . . . . . . . . . . . . 21
2.4.3 Base Extra Method . . . . . . . . . . . . . . . . . . . . . . 21
2.5 A Functional Approach to Water Pricing . . . . . . . . . . . . . . 22

CHAPTER 3: Production and Costs . . . . . . . . . . . . . . . . . . . . . 25

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.2 Production and Costs in Urban Water Supply . . . . . . . . . . . 25
3.3 Functional Modelling of Production and Cost . . . . . . . . . . . 27

iii
 3.4 Theory of the Firm . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.4.1 The Constant Elasticity of Substitution Production Function 30
3.4.2 The Constant Returns to Scale Minimum Cost Function . 32
3.4.3 Nonconstant Returns to Scale Minimum Cost Function . . 34
3.5 The Translog Cost Functional Form . . . . . . . . . . . . . . . . . 37
3.6 Estimation of the Production and Cost Functions . . . . . . . . . 38
3.6.1 Issues in Estimation . . . . . . . . . . . . . . . . . . . . . 38
3.6.2 Specification of the Constant Returns to Scale Minimum
Cost Function . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.6.3 Specification of the Variable Returns to Scale Minimum
Cost Function . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.6.4 Translog Cost Function . . . . . . . . . . . . . . . . . . . . 42
3.6.5 Panel Data Estimation . . . . . . . . . . . . . . . . . . . . 43
3.7 Chapter Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . 44

CHAPTER 4: Household Demand . . . . . . . . . . . . . . . . . . . . . . 45

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.2 Consumer Optimisation Problems . . . . . . . . . . . . . . . . . . 46
4.2.1 Demand Functions . . . . . . . . . . . . . . . . . . . . . . 47
4.2.2 Applied Demand Analysis and the Almost Ideal Demand
System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.3 Modelling Demand subject to a Block Rate Tariff . . . . . . . . . 51
4.3.1 The IBRT Demand Function . . . . . . . . . . . . . . . . . 52
4.3.2 Estimation of the IBRT Demand Function . . . . . . . . . 56
4.3.2.1 Least Squares . . . . . . . . . . . . . . . . . . . . . . . . 56
4.3.2.2 The Two Error Specification . . . . . . . . . . . . . . . . 57
4.3.2.3 Measurement Error . . . . . . . . . . . . . . . . . . . . . 59
4.4 Maximum Likelihood Estimation of the Two Error Model . . . . . 60
4.4.1 The Likelihood Function . . . . . . . . . . . . . . . . . . . 61
4.4.2 Extending the Model with more than Two Blocks . . . . . 63
4.4.2.1 Multiple Choice Method . . . . . . . . . . . . . . . . . . 63
4.4.2.2 Binary Choice Method . . . . . . . . . . . . . . . . . . . 65
4.5 Chapter Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . 66

CHAPTER 5: Welfare and Optimal Pricing . . . . . . . . . . . . . . . . . 67

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
5.2 Measurement of Welfare . . . . . . . . . . . . . . . . . . . . . . . 67
5.3 Measurement of Welfare Change . . . . . . . . . . . . . . . . . . . 70
5.4 Welfare and Economies of Scale . . . . . . . . . . . . . . . . . . . 73
5.4.1 Marginal Cost, Average Cost, and Economies of Scale . . . 73
5.4.2 Cost Recovery under Marginal Cost Pricing . . . . . . . . 75
5.4.3 Equity in Ramsey Pricing . . . . . . . . . . . . . . . . . . 75
5.5 Estimation Issues related to Welfare Models . . . . . . . . . . . . 78

iv
CHAPTER 6: First Case Study
Urban Water Services in Victoria
Part A - Cost . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
6.1 The Supply of Urban Water in Victoria . . . . . . . . . . . . . . . 81
6.2 The Cost Data Set for Victorian Water Businesses . . . . . . . . . 83
6.2.1 Output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
6.2.2 Revenue . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
6.2.3 Operational Costs and Factor Prices . . . . . . . . . . . . 85
6.2.3.1 Capital Costs . . . . . . . . . . . . . . . . . . . . . . . . 86
6.2.3.2 Labour . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
6.2.3.3 Service Costs and Technology . . . . . . . . . . . . . . . 87
6.3 Data Description . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
6.4 Estimation of Two Factor Production and Cost Models . . . . . . 90
6.4.1 Two Factor CRS Production Function . . . . . . . . . . . 90
6.4.2 Two Factor Cost Function with CRS Production . . . . . 91
6.4.3 Two Factor Cost Function with Variable Returns to Scale 94
6.5 Discussion of Results . . . . . . . . . . . . . . . . . . . . . . . . . 96
6.5.1 Testing for Constant Returns to Scale . . . . . . . . . . . . 98
6.6 Applying a Translog Cost Function to Analyse the Impact of Service
Costs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103

CHAPTER 7: First Case Study

Urban Water Services in Victoria
Part B - Marginal Cost and Welfare . . . . . . . . . . . . . . . . . . . . 106
7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
7.2 Estimating Marginal Cost . . . . . . . . . . . . . . . . . . . . . . 107
7.2.1 Constant Returns and Variable Returns Models . . . . . . 107
7.2.2 Partial Equilibrium Marginal Cost . . . . . . . . . . . . . 108
7.2.3 Translog Function Marginal Cost . . . . . . . . . . . . . . 108
7.3 Discussion of Results . . . . . . . . . . . . . . . . . . . . . . . . . 110
7.3.1 Data Aggregation and Sensitivity Analysis . . . . . . . . . 113
7.4 Measurement of Deadweight Loss and Required Subsidy . . . . . 116
7.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119

CHAPTER 8: Second Case Study

Urban Water Services in Manila
Part A - Cost . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
8.1 The Privatisation of Water Services in Manila . . . . . . . . . . . 122
8.2 Production and Costs . . . . . . . . . . . . . . . . . . . . . . . . . 128
8.2.1 Output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
8.2.2 Operational Costs and Factor Prices . . . . . . . . . . . . 131
8.2.2.1 Overhead Costs . . . . . . . . . . . . . . . . . . . . . . . 131
8.2.2.2 Labour . . . . . . . . . . . . . . . . . . . . . . . . . . . 131

v
 8.2.2.3 Cost of Capital . . . . . . . . . . . . . . . . . . . . . . . 132
8.2.2.4 Energy Costs and Technology . . . . . . . . . . . . . . . 134
8.2.2.5 Environmental Costs . . . . . . . . . . . . . . . . . . . . 134
8.2.2.6 Foreign Exchange Losses and Gains . . . . . . . . . . . . 134
8.3 Data Description . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
8.4 Estimation of the Minimum Cost Models . . . . . . . . . . . . . . 137
8.4.1 Two Factor Cost Function with CRS Production . . . . . 138
8.4.2 Two Factor Cost Function with Variable Returns to Scale 139
8.5 Discussion of Results . . . . . . . . . . . . . . . . . . . . . . . . . 140
8.6 Testing for Constant Returns to Scale . . . . . . . . . . . . . . . . 143

CHAPTER 9: Second Case Study

Urban Water Services in Manila
Part B - Demand . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
9.1 Urban Water Consumption and Prices in Manila . . . . . . . . . . 145
9.1.1 Sources of Data . . . . . . . . . . . . . . . . . . . . . . . . 145
9.1.1.1 The 2000 Census of Population and Housing . . . . . . . 146
9.1.1.2 The 2000 Family Income and Expenditure Survey . . . . 146
9.1.2 Household Sources of Water . . . . . . . . . . . . . . . . . 147
9.1.3 Water Prices and Household Income . . . . . . . . . . . . 148
9.2 Consumption Analysis using Linear Methods . . . . . . . . . . . . 151
9.2.1 Data Preparation and Description . . . . . . . . . . . . . . 152
9.2.2 Estimation of the Linear Demand Function . . . . . . . . . 156
9.3 Consumption Analysis using Maximum Likelihood Methods . . . 159
9.3.1 Data Preparation and Description . . . . . . . . . . . . . . 159
9.3.1.1 Calculation of Water Consumption . . . . . . . . . . . . 160
9.3.1.2 Descriptive Statistics . . . . . . . . . . . . . . . . . . . . 164
9.3.2 Estimation Results . . . . . . . . . . . . . . . . . . . . . . 164

CHAPTER 10: Second Case Study

Urban Water Services in Manila
Part C - Marginal Cost and Welfare . . . . . . . . . . . . . . . . . . . . 169
10.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
10.2 Marginal Cost Estimation . . . . . . . . . . . . . . . . . . . . . . 170
10.2.1 Marginal System Cost . . . . . . . . . . . . . . . . . . . . 171
10.2.2 Marginal Cost in Per Connection Terms . . . . . . . . . . 171
10.2.3 Marginal Cost at a Partial Equilibrium . . . . . . . . . . . 172
10.3 Welfare Effects at Average Price Levels . . . . . . . . . . . . . . . 174
10.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176

CHAPTER 11: Performance Based Pricing . . . . . . . . . . . . . . . . . 178

11.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
11.2 Review of Case Studies . . . . . . . . . . . . . . . . . . . . . . . . 179
11.3 Price and Sustainable Water Management . . . . . . . . . . . . . 182

vi
 11.4 Cost Minimisation in the Supply of Urban Water . . . . . . . . . 184
11.5 A Performance Based Approach to Urban Water Pricing . . . . . 185
11.5.1 Stochastic Frontier Analysis . . . . . . . . . . . . . . . . . 187

CHAPTER 12: Third Case Study

Performance Based Pricing:
A Frontier Analysis of Asian and Australian Water Utilities . . . . . . 189
12.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
12.2 Source of Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
12.3 Data Description . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
12.4 Estimation Results . . . . . . . . . . . . . . . . . . . . . . . . . . 195
12.5 Discussion of Results . . . . . . . . . . . . . . . . . . . . . . . . . 196
12.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200

CHAPTER 13: Conclusion and Directions for Future Research . . . . . . . 206

13.1 Resume of this Research . . . . . . . . . . . . . . . . . . . . . . . 206
13.2 Originality of this Contribution . . . . . . . . . . . . . . . . . . . 209
13.3 Urban Water Pricing Policy . . . . . . . . . . . . . . . . . . . . . 210
13.3.1 Tariff Design and Variable Unit Pricing . . . . . . . . . . . 211
13.3.2 Efficiency and Competition . . . . . . . . . . . . . . . . . 214
13.3.3 Agricultural Water Use and Trade . . . . . . . . . . . . . . 215
13.3.4 Data Standardisation . . . . . . . . . . . . . . . . . . . . . 215
13.3.5 Other Implications of Pricing Policy . . . . . . . . . . . . 216
13.4 Future Research Directions . . . . . . . . . . . . . . . . . . . . . . 218
13.5 Closing Comments . . . . . . . . . . . . . . . . . . . . . . . . . . 219

APPENDIX A: Properties of the Variable Returns to Scale Cost Function 221

APPENDIX B: Transformation of the Two Error Model Likelihood Function 224

B.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
B.2 Segment Likelihood . . . . . . . . . . . . . . . . . . . . . . . . . . 224
B.3 Kink Likelihood . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226
B.4 Identities used in Transformations . . . . . . . . . . . . . . . . . . 227
B.4.1 Properties of the Error Variables: . . . . . . . . . . . . . . 227
B.4.2 Properties of the Standardised Error Variables: . . . . . . 228

BIBLIOGRAPHY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229

vii
 FIGURES

4.1 Nonconvex Budget Constraint . . . . . . . . . . . . . . . . . . . . 54

4.2 Convex Budget Constraint . . . . . . . . . . . . . . . . . . . . . . 55
4.3 Distribution of . . . . . . . . . . . . . . . . . . . . . . . . . . . 59

5.1 Maximisation of Total Net Benefit . . . . . . . . . . . . . . . . . . 69

5.2 Deadweight Loss from Inefficient Pricing . . . . . . . . . . . . . . 71
5.3 Deadweight Loss under Constant Marginal Cost . . . . . . . . . . 72
5.4 Deadweight Loss under Decreasing Marginal Cost . . . . . . . . . 72

10.1 Partial Equilibrium for the Manila Concessions . . . . . . . . . . 176

viii
 TABLES

6.1 Cost and Revenue Items Identified in Financial Statements . . . 84

6.2 Victorian Regional Water Businesses: Summary Statistics 2002-
2005 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
6.3 Victorian Metropolitan Water Businesses: Summary Statistics 2002-
2005 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
6.4 CRS Production Function . . . . . . . . . . . . . . . . . . . . . . 92
6.5 CRS Cost Function . . . . . . . . . . . . . . . . . . . . . . . . . . 93
6.6 Variable RTS Cost Function: . . . . . . . . . . . . . . . . . . . . 95
6.7 Summary of Estimation Results: Victorian Cost Data . . . . . . . 99
6.8 F Tests for Model Significance . . . . . . . . . . . . . . . . . . . . 100
6.9 Ramsey RESET Tests for Model Specification Error . . . . . . . . 101
6.10 Hypothesis Tests for Constant Returns to Scale . . . . . . . . . . 102
6.11 Share Equation Estimates: Translog Cost Function . . . . . . . . 105
6.12 Estimates of Elasticities: Translog Cost Function . . . . . . . . . 105

7.1 Marginal Cost Results: Victorian Water Authorities . . . . . . . . 111

7.2 Sample Means of Parameters: Victorian Water Authorities . . . . 112
7.3 Sensitivity Analysis of Marginal Cost Estimates: Variable Returns
Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
7.4 Welfare Measurements: Variable Returns Model . . . . . . . . . . 118

8.1 Metro Manila Water Tariff and Production 1996-2003 . . . . . . 125

8.2 Manila Cost Data: Cost Items Identified in Financial Statements 130
8.3 Manila Cost Data: Summary Statistics . . . . . . . . . . . . . . . 136
8.4 Two Factor Cost Function with CRS Production . . . . . . . . . . 139
8.5 Two Factor Cost Function with Variable RTS Production . . . . . 140
8.6 Summary of Estimation Results: Manila Cost Data . . . . . . . . 142
8.7 F Tests for Model Significance . . . . . . . . . . . . . . . . . . . . 142

ix
8.8 Ramsey RESET Tests for Model Specification Error . . . . . . . . 143
8.9 Hypothesis Tests for Constant Returns to Scale . . . . . . . . . . 144

9.1 Source of Household Water in Manila NCR . . . . . . . . . . . . 148

9.2 Inocencio Study of Household Water Source . . . . . . . . . . . . 150
9.3 Waters Share of Household Budget (%) . . . . . . . . . . . . . . 152
9.4 Weighted Average Domestic Water Price (Pesos per m3 ) . . . . . 154
9.5 NCR Water Consumption, Pricing, and Expenditure Data Set:
Summary Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . 155
9.6 Estimates of Demand Function - NCR Water Consumption . . . . 157
9.7 Price and Income Elasticities - NCR Water Consumption . . . . . 158
9.8 Tariffs for Domestic Water Use - Manila 2000 . . . . . . . . . . . 161
9.9 MWSS Water Consumption Data Set : Summary Statistics . . . . 163
9.10 Estimates of Demand Function - MWSS Connected Households . 167
9.11 Price and Income Elasticities - MWSS Connected Households . . 168
9.12 Likelihood Ratio Test . . . . . . . . . . . . . . . . . . . . . . . . . 168

10.1 Summary of Marginal Cost Estimates: Manila Water Concessions 173

10.2 Sample Means of Parameters: Manila Water Concessionaires . . . 175
10.3 Welfare Measurements: Variable Returns Model . . . . . . . . . . 175

12.1 IBNET Performance Indicators: Summary Statistics 2000-2005 . 193

12.2 IBNET Performance Indicators: Sample Means Comparison 2000-
2005 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
12.3 IBNET Country Codes . . . . . . . . . . . . . . . . . . . . . . . 195
12.4 Stochastic Frontier Analysis: Cost Function . . . . . . . . . . . . 197
12.5 Stochastic Frontier Analysis: Cost Efficiency Scores . . . . . . . . 201

x
 ABBREVIATIONS

2SLS Two-Stage Least Squares

ADB Asian Development Bank

ADERASA Association of Water and Sanitation Regulatory Entities of the Americas

ADR Appropriate Discount Rate (Phil.1 )

AEPA Accelerated Extraordinary Price Adjustment (Phil.)

AWWA American Water Works Association

AUD Australian Dollar

BEA Bureau of Economic Analysis (USA)

BOOT Build Own Operate Transfer (Vic.2 )

CAPEX Capital Expenditure

CERA Currency Exchange Rate Adjustment (Phil.)

CES Constant Elasticity of Substitution

CPI Consumer Price Index

CRS Constant Returns to Scale

DCRA Debt Capital and Restructuring Agreement (Phil.)

DEA Data Envelopment Analysis

EEPSEA Economy and Environment Program for Southeast Asia (Canada)

ESC Essential Services Commission (Vic.)

FCDA Foreign Currency Devaluation Adjustment (Phil.)

1
Philippines
2
Victoria, Australia

xi
FIES Family Income and Expenditure Survey (Phil.)
FTE Full Time Equivalent

HPE Heterogeneous Preferences Error

IBRD International Bank for Reconstruction and Development

IBRT Increasing Block Rate Tariff

IDA International Development Agency

IFC International Finance Corporation

IPART Independent Pricing and Regulatory Tribunal (NSW)

KL Kilolitres

KLM Kilolitres per Month

LIBOR London Interbank Overnight Rate

ML Megalitres

MLD Megalitres per Day

MLE Maximum Likelihood Estimation

MWCI Manila Water Company Inc. (Phil.)

MWSI Maynilad Water Supply Inc. (Phil.)

MWSS Metropolitan Water and Sewerage Service (Phil.)

MWSSRO Metropolitan Water and Sewerage Service - Regulatory Office

NCR National Capital Region (Phil.)

NGO Non-Government Organisation

NRW Non-Revenue Water

NSO National Survey Office (Phil.)

NWC National Water Commission (Australia)

OECD Organisation for Economic Cooperation and Development

OFWAT Office of Water Services (UK)

OLS Ordinary Least Squares

xii
PAWS Public Assessment of Water Services (Phil.)
PhP Philippine Peso

PPE Property Plant and Equipment

RESET Regression Specification Error Test

RTS Returns to Scale

SEAWUN Southeast Asian Water Utilities Network

SFA Stochastic Cost Frontier

SPR Service Performance Report (Phil.)

SUR Seemingly Unrelated Regressions

UATP Umiray Angat Transbasin Project (Phil.)

USD US. Dollar

VAT Value Added Tax (Phil.)

VRS Variable Returns to Scale

WACC Weighted Average Cost of Capital

WIRO Water Industry Regulatory Order (Vic.)

WUP Water Utility Partnership (Africa)

xiii
 Marginal Cost Water Pricing: Welfare Effects and Policy Implications

using Minimum Cost and Benchmarking Models,

with Case Studies from Australia and Asia

ABSTRACT

Recent studies in water management policy point to insufficient recognition of

water as a scarce commodity and the failure of pricing policies to account for the
full economic costs of its production and supply. These costs include opportunity
costs related to alternative uses of water; user costs associated with managing a
scarce resource; and costs of externalities such as ground water depletion, pollution

of waterways, and greenhouse gas emissions. Existing cost recovery based pricing
policies may lead to inefficiencies such as excess consumption, under-investment
in water infrastructure, and unnecessary subsidisation.
Water scarcity can be managed in several ways. We can increase supply by
investment in additional harvesting capabilities or new technologies such as de-

salination; we can constrain consumption so that existing supplies last longer; or

we can use water in more efficient ways. As a short term measure, most countries
adopt water restrictions when supplies are at critical levels. In the future, as urban
population growth continues, harvesting of storm water and reuse of grey water

may become part of a sustainable water management strategy. Water trading can
be used to move water to where the marginal benefits are highest. Considerable
water savings are possible through the use of more efficient industrial and domes-
tic appliances. There is evidence in some countries that higher water tariffs have

reduced consumption and promoted awareness of conservation. If we accept that

xiv
water is an economic good, then we need to understand the costs related to its
production, the patterns of its use, and the benefits received by different users.

This thesis is an examination of theoretical and applied aspects of urban water

pricing based on analysis of cost, demand, and welfare. We present theoretical
models of cost that include economies of scale as a parameter, and a model of water
demand by households with heterogeneous preferences. We determine marginal

cost at the efficient level of output based on a partial equilibrium of supply and
demand. We also show that when water is produced with increasing returns to
scale, the efficient price will be insufficient to recover all costs, and therefore a form
of second best pricing is required. We contrast conventional notions about water

suppliers being cost minimisers with an alternative frontier model of cost efficiency.
Two case studies examine the provision of water services under different forms of
ownership. The first case study examines the provision of water to domestic
households in the state of Victoria, Australia. The second case study examines
the supply of water to the residents of Manila, one of the worlds largest cities

that privatised its water service in 1997 under a form of concession agreement.
A third case study derives an efficient cost frontier for a sample of water utilities
from Asia and Australia and proposes a form of best practice pricing. The thesis
concludes with a summary of the main results and policy conclusions, and ideas

for future research.

xv
 THESIS DECLARATION

This work contains no material which has been accepted for the award of any
other degree or diploma in any university or other tertiary institution and, to the
best of my knowledge and belief, contains no material previously published or
written by another person, except where due reference has been made in the text.

I give consent to this copy of my thesis, when deposited in the University

Library, being made available in all forms of media, now or hereafter known.

David Altmann
Adelaide University Student ID. 1117938

xvi
 ACKNOWLEDGMENTS

Over the course of this project I have been fortunate to have been helped by
a number of people in both the Philippines and Australia. In the Philippines,
I would like to acknowledge the assistance of staff of the Metropolitan Water
and Sewerage Service including the acting chief regulator, Colonel Angel Efren

Agustin, also Randolph Sakai, Rosenddo Alegre, and Charlie Espallardo. Staff
of the National Statistics Office helped with the census information including the
Chief of the Census Planning and Operations Division, Mercedita Tia and her
staff; Emma Fabian, Sol Vergara, Gene Lowrica, and Vilex Malumay. Early in

the research I consulted with several staff of the University of the Philippines
and Ateneo De Manila University. These included Dr. Erniel Barrios and Dr
Rosalina Tan. I was also fortunate to have met and been helped by Dr. Cristina
David, Dr. Arlene Inocencio and Jenny Furliguton of the Philippines Institute for
Development Studies.

In Australia, I would also like to acknowledge the assistance of staff from a

number of Victorian water businesses including Rebecca McIntyre (Barwon Wa-
ter), Mary Connelly-Gale (Goulburn Valley Water), Pam Fitzpatrick and Keith
Thomson (Lower Murray Water), Karen Smith and Paul Deveraux (Central High-

lands Water) and Helen Friend (Grampians Wimmera-Mallee Water); also from
the Essential Services Commission, Robin Keely. Professor Mike Young from the
University of Adelaide has provided helpful guidance and further afield Dr. Jon
Isham from Middlebury College in the USA has assisted with data sets.

xvii
 Finally, I would like to acknowledge my supervisors, Dr Eran Binenbaum, Dr
John Hatch, and Dr Ralph Bayer for their comments and assistance over the

period this work has been undertaken.

xviii
 CHAPTER 1

Introduction

The sustainable use of water is one of most important challenges of our time.
Urban water supplies are under considerable pressure in a number of the worlds
major cities as a result of ageing infrastructure, a decline in investment, increased
demand from population growth, and the migration of rural workers to cities. The

drought in Australia has placed severe stress on water supplies, for both agriculture
and urban dwellers. Storage dams in our major cities are at record low levels.
Warragamba dam, Sydneys main supply, is at 33.9% of capacity; the Thomson
reservoir in Melbourne is at 18.2% of capacity1 . Queensland is operating under
level five water restrictions, the highest ever applied, where residential users can

be fined for excess consumption. The Murray Darling Basin, upon which Adelaide
relies for most of its water, has received inflows in the past year that are only 60%
of the previous low level set in 1983 (Australian Broadcasting Commission, 2007).
The problems are not limited to Australia. Of the worlds total population, an

estimated twenty percent do not have access to potable water (United Nations,
2006). Between 2000 and 2030, most population growth is expected to occur
within the urban areas of less developed countries. This is expected to place urban
supplies under even greater pressure. In response to the crisis, water management

and use have moved to the forefront of policy debate, water is becoming more
valued, and innovative solutions are being sought and evaluated.
1
First quarter 2007 figures supplied by Sydney Catchment Authority and Melbourne Water
respectively.

1
 Supply augmentation, construction of new infrastructure to increase harvest
and storage resources, has been the usual approach employed to manage growth in

water consumption. Governments have become reluctant to adopt this approach

in the recent past because it requires debt funding of large capital investments, and
suitably located sites have been given over to urban or agricultural development.
When the need to carry out new investment is unavoidable however, because of

population growth, reduced inflows, or ageing infrastructure; it is likely to result

in considerably higher marginal costs than the existing supply. If marginal costs
increase, either the level of subsidisation needs to increase, or long run prices and
allocations will adjust so that the marginal benefit for different classes of users

are equal. Potentially, urban and agricultural users of water will compete for the
same resource (Saleth and Dinar, 1997).
Demand management is acknowledged as an important part of a mix of strate-
gies that will be required to achieve sustainability in water resource management.
This can take several forms. Firstly, in Australia, we have experienced demand

management in the form of water restrictions over many years. In the past, these
were seasonal, however, restrictions have now become permanent in a number
of states due to drought conditions. Given the considerable penalties for failure
to comply, water restrictions do reduce consumption when brought into force.

Secondly, innovative pricing mechanisms such as block rate pricing are becoming
increasingly common. An increasing block rate will reduce overall consumption
of water, because once basic household needs are satisfied, consumption becomes
more sensitive to price. In the presence of a well designed block rate pricing tariff,

consumers are less willing to pay for high water use activities such as watering
of gardens or car washing. Thirdly, demand management may take the form of
seasonal adjustments to price, for example to cope with the increase in demand
in summer. This approach has yet to be adopted in Australia.

2
 Recognition that water is an economic good is a precondition for implemen-
tation of sustainable strategies such as demand management. This means that

water prices need to be determined in accordance with economic principles rather

than those that focus on cost recovery. Accounting for the costs of water supply
should include opportunity and environmental costs, in addition to operational
and investment costs. Market based marginal cost pricing needs to be considered

as a replacement for the established practice of rate of return pricing whereby a

utility prices its service according to average cost plus a minimum rate of return
on capital.
The need for change is already accepted by many water utilities and their own-

ers. Government owners are more aware of the need for an economic return from
water supply investment that accounts for the long run marginal costs. Market
based mechanisms are already in place that allow water to move between and
across sectors, from rural to rural and rural to urban, to those users that value
the resource more. This promotes conservation among users and establishes clear

market driven pricing signals. With more transparent data about prices and de-
mand, governments and the private sector can be better informed about long term
investment opportunities.
The changes in thinking about water, coupled with the knowledge that gov-

ernments have in many cases been unsuccessful in managing the resource, have
acquired large debt burdens, and often have been unable to move beyond least
cost recovery pricing mechanisms, has led to privatisation of water supply op-
erations in many countries. The reaction to the price increases that invariably

follow, or the lack of service improvement, has been negative, indeed sometimes
violent (Forero, 2005). Opinions regarding the success of private sector manage-
ment of water service are polarised, as are the outcomes. Some countries have
been successful, in others renationalisation appears imminent (Quiggin, 2002).

3
 This thesis is a microeconomic analysis of the provision of water services in
an urban context. Our primary research objective is to examine marginal cost

water pricing by asking the following questions. How is the marginal cost of
water determined? How is the price of water established using marginal cost as
the basis? How do current water prices compare with estimated price based on
marginal cost? What is the welfare impact of current water pricing relative to

a proposed marginal cost based price? What are the main policy issues that we
need to address in a debate on water pricing - particularly one based on marginal
cost? In trying to provide answers, we will use the tools of applied microeconomic
and econometric analysis and focus only on the provision of urban water services;

including harvesting, production, and distribution. We hope that this work will
contribute to our understanding of the variables and the relationships that are
significant, the policy directions that will maximise benefits for society, and where
our future efforts in research and understanding ought to be focused.
The novelty of the approach rests with three key features of this work. The

first is extensive use of functional forms to describe the key variables: cost, de-
mand, and welfare. This is in contrast to industry practice which relies on present
value techniques to derive scalar quantities of the key variables. The second is
the recognition that water utilities often operate with increasing returns to scale

and therefore the efficient economic solution based on marginal cost pricing means
that the utility does not recoup its costs. We therefore consider the use of Ramsey
pricing. The third feature is an acceptance that utilities may not behave as cost
minimisers in all cases, particularly if operating under rate of return regulation.

Therefore we have proposed an alternate pricing mechanism based on use of effi-

ciency frontiers. These areas have been dealt with independently but seldom in
a unified manner. A fuller discussion of the original aspects of this work will be
reserved until the thesis conclusion.

4
 This thesis consists of thirteen chapters. Chapter 1 is this introduction that
sets out the motivation for this research and an outline of the document structure.

Chapter 2 is a description of existing water pricing practices and the regulatory

contexts that determine water prices in many parts of the world. In this chapter
we examine rate of return regulation, the dominant method used to price water.
We also examine marginal cost pricing, price caps and other forms of regulation

such as yardstick regulation that are applicable to water utilities. Chapters 3, 4,

and 5 are devoted to theory and empirical issues. Chapter 3 is primarily devoted to
description of a cost function that includes a parameter for economies of scale and
that can be used to accommodate increasing, decreasing or constant marginal cost.

We also outline the use of translog cost functions under assumptions of constant
returns to scale production technology. Chapter 4 presents two alternative models
for estimation of domestic water demand: the linear almost ideal demand system,
and a two error heterogeneous preferences model for use in block rate water pricing.
This chapter discusses empirical issues surrounding use of the two error model and

the use of maximum likelihood estimation for estimation of the parameters of the
demand function. In Chapter 5, we consider different ways of determining the
marginal cost including an efficient allocation that maximises total net benefit for
consumers and producers. We examine the welfare outcomes related to pricing at

below marginal cost based on whether marginal costs are increasing, decreasing
or constant. We also discuss issues of subsidisation and equity under second best
(Ramsey) prices and the determination of a partial equilibrium when demand
elasticity is known but the parameters of the complete demand function are not

known.
Chapters 6 to 10 constitute the main body of applied work and are centered
around two empirical case studies. The theoretical and methodological material of
Chapters 3 to 5 are applied to the supply of urban water in one state of Australia

5
and the capital city of the Philippines.
Chapters 6 and 7 contain a case study based on cost and demand data from

the seventeen water businesses that supply water to domestic households across
the state of Victoria. These businesses are owned by the state government but
operate as corporations. Based on an unbalanced panel data set of accounting
data, we estimate cost functions including constant returns to scale cost, variable

returns to scale (scale economies), and constant returns translog cost function.
The econometric techniques used include ordinary least squares and seemingly
unrelated regression estimation. We determine the marginal cost based on a par-
tial equilibrium and estimate the welfare losses associated with current prices, and

estimate the subsidy still required under marginal cost pricing.

Chapters 8 to 10 contain a case study that focuses on the recently privatised
supply of water to residential users in Manila in the Philippines. We examine
the historical events surrounding the privatisation that occurred in 1997 when
responsibility for operation and maintenance of the water supply was passed to

two private companies under a 25 year concession agreement. Following this we

carry out a cost and demand analysis similar to the preceding case study. In
this case study we estimate the demand function based on data from a national
household expenditure survey and, using this demand function, we estimate the

welfare losses associated with current prices.

Although this work is not a comparative study, in Chapter 11 we identify and
explain some common and contrasting results based on the two case studies. From
this analysis we identify weaknesses in the marginal cost approach particularly

in respect of the use of cost minimisation as the basic theoretical assumption.

We then present an alternative approach based on the use of a stochastic cost
frontier to model firms. This approach is gaining recognition in benchmarking and
performance based utility pricing. Chapter 12 is a third case study that presents

6
an application of performance modelling using a stochastic frontier model and a
benchmarking data set originating from the World Bank. This case study suggests

various ways in which price might be determined using this kind of performance
based approach.
Chapter 12 concludes the thesis with a summary of the main results and
presents a number of areas in which policy might be focused based on the ev-

idence from the case studies and some stylised facts that have emerged from the
study. Finally, we identify some new research questions and the proposed direc-
tion of future research in this area. A bibliography of cited references appears at
the very end of the document.

7
 CHAPTER 2

Urban Water Pricing

2.1 Introduction

The objective of this chapter is to review theoretical and conventional urban

water pricing practice, and to establish both the motivation for use of marginal
cost pricing and an understanding of the limitations of its use. We start with
an overview of theoretical developments in utility pricing in general and water
pricing in particular, before discussing the role of regulation, and the relationship
between the form of regulation and the determination of price. We next review

two alternative approaches to water pricing: rate of return pricing and marginal
cost pricing. We will outline the case in support of marginal cost pricing and
examine some of the limitations of its use. The following three chapters will then
focus on development of cost, demand, and welfare models applicable to marginal

cost pricing of water services.

2.2 Theoretical Foundations of Utility Pricing - First and Second Best Solutions

The economic basis of water pricing is that the supplier must be able to recover

all of its costs including operations and maintenance, investment, and social costs
including externalities. The economically efficient basis for pricing would be to set
the price at the level of long run marginal cost so that net surplus is maximised. We
will review some of the practical difficulties of this approach later in this chapter,

8
but for now we concentrate only on the theoretical approaches that evolved during
the twentieth century. At the heart of the problem of utility pricing (or of any

network industry that exhibits increasing returns to scale) is that marginal cost
pricing will result in a unit price that is less than average cost, therefore the utility
will not generate sufficient revenue to cover costs. First and Second Best solutions
differ mainly in how this shortfall is recouped (Harris, Tate, and Renzetti, 2002).

First Best pricing solutions rely on some form of lump sum transfer to make
up the loss. Hotelling (1938) proposed the use of subsidies and taxation. This
approach was criticised by Coase (1946) who considered subsidies to be distor-
tionary. For example, a small community that is able to supply its own needs will

most likely switch to a lower cost subsidised public utility despite the full cost of
provision being higher than if they continued to self supply. The solution proposed
by Coase entailed the use of a fixed charge covering the cost of network connection
and a volumetric charge. The volumetric charge would be set at the marginal cost
of supply, while the fixed charge would be set to make up the revenue shortfall.

Subsequently, Vickrey (1955) pointed out that volume and distribution were two
complementary goods - both with variable costs (volume decreasing marginal cost,
while distribution was increasing moving to the network extremities). Residual
costs could be recouped as fixed network access charges. Vickrey also argued that

where consumers had the option of opting out of the supply, or not connecting at
all, there was a degree of cross-price elasticity between volume and distribution.
Second Best or Quasi-Optimal Solutions are characterised by the use of price
discrimination to recover costs. Of these Ramsey pricing and Pareto Superior

Non-Linear Outlay Schedules are the most common. A Ramsey price is set at
the welfare maximising level of output using price discrimination based on the
demand elasticity of different user classes. This is calculated as a markup from
the marginal cost - the Ramsey number (R). In the simplest case where there

9
is only one good produced and the cross price elasticity of demand among user
classes is zero, the Ramsey number is the percentage deviation from marginal cost

weighted by the demand elasticity:

 
pi mi
Ri = i
pi

where for the ith class of user, pi is the volumetric charge for the good, mi is its
marginal cost, and i is the price elasticity of demand. Ramsey numbers are con-
strained to the interval 0 < R < 1, with 1 corresponding to a price discriminating
monopolist who recovers all costs, while 0 is perfect competition.

Ramsey pricing of water presents a difficulty in terms of equity because lower

income consumers with inelastic demands will pay a higher unit price relative to
those whose demand is more elastic such as higher income residential or industrial
consumers. Pareto Superior Non-Linear Outlay Schedules (more simply declining

block rate tariffs or volume discounts) were developed as a means of overcoming

these problems. The introduction of one or more blocks with declining unit price
results in an increase in both firm profit and consumer surplus, hence of net
welfare. Willig (1978) showed that for any case where average cost was greater

than the marginal cost, there exists a declining block rate structure that Pareto
dominates the average cost price. The Pareto Superior solution stands in contrast
to the case of increasing marginal cost of production caused, for example, by water
scarcity. In this case the efficient solution is to apply an increasing block rate
tariff that approximates the marginal cost curve where marginal cost is in excess

of average cost (where there are decreasing returns to scale) and revert to the cost
recovery methods where this is not the case. At higher levels of consumption,
demand is more elastic so that the use of increasing block rates can be effective
in conserving water.

10
 In summary, from a theoretical viewpoint, the problem of utility pricing and
water pricing in particular arise because the network structure leads to a form of

monopolistic production characterised by increasing returns to scale. Under these

conditions, if price is set at marginal cost, some form of transfer is required to
make up the shortfall. Otherwise the problem becomes one of determining the
optimum price level that is Pareto efficient for different classes of user. In the

next section we discuss the regulatory environment within which this price setting
occurs.

2.3 The Regulatory Environment

Natural monopolies arise in network industries such as water supply because

high capital requirements create barriers to entry and the entrenched owner can
produce under conditions of economies of scale and scope - therefore at a lower cost
than the cost of multiple individual firms (sub-additivity). Regulation evolved in
the nineteenth century to protect consumers from the potential abuse of monopoly

powers while preserving the interests of investors. Utilities initially operated un-
der private ownership, but where the interests of consumers and owners was in
conflict, nationalisation took place (Newbery, 1998). Gas and water industries
were renationalised in the United Kingdom in 1949 and 1973 respectively fol-

lowing excessive rent seeking by the private operators of those services. Similar
events took place in Australia after the Second World War; for example, the South
Australian Electricity Trust was nationalised in 1946, and the airline Qantas in
1947. The return of nationalised assets to private ownership in the latter part of

the twentieth century (in the UK this occurred over a period 1984-1994, and in
Australia in the 1990s) was in response to a widely held perception that years of
public ownership had distorted prices and created inefficiencies, often as a result
of repeated political interference (Saal and Parker, 2004). Private industry was

11
considered a better vehicle to deliver cost efficiencies, service improvements and
investment. In the developing parts of the world, the IMF and World Bank have

been strong advocates of privatisation as a means of debt reduction and increas-

ing access to capital markets for funding new infrastructure projects (Magdahl,
Srreimd, Christensen, Preston, Kronen, and Berg, 2006). Corresponding with the
wave of privatisation that commenced over two decades ago, a considerable body

of literature has evolved in the area of regulation and pricing of utilities, includ-
ing water supply. This work was driven by the need for an expanded regulatory
role in those countries where major privatisations had occurred, in particular the
United Kingdom and the United States. Applied research focused on the impact

of ownership on productivity and efficiency, and the extent to which consumers

have benefited from changes in ownership. The studies of urban water supply
undertaken since this period including, for example, Bhattacharyya, Parker, and
Raffiee (1994); Bruggink (1982); Feigenbaum and Teeples (1983); Renzetti (1999);
Saal and Parker (2004) aim to determine differences, if any, in cost efficiency or

productivity by firms under private or public ownership. Often the applied results
are inconclusive and sensitive to specification. A consensus is emerging from the
empirical literature however, that it is the presence of competition that leads to
cost efficiencies, rather than ownership per se. This debate is by no means over

and remains an active area of research - nor is it the central theme of this study,
but it has produced a great amount of useful applied research in the area of cost
models and efficiency measurement.
One could argue that, using some form of Ramsey pricing, price determination

is a mechanical process once long run marginal costs are understood and quanti-
fied. However, regulators have been reluctant to adopt this method because long
run marginal costs are by definition forward looking and it is simpler to base price
on historical data, adjusted for projected growth in demand and costs. Further-

12
more long run marginal costs should include costs that are related to scarcity,
negative externalities such as greenhouse emissions, and the positive externalities

arising from quality potable water. Much of the research in this area ignores
these factors because data limitations mean these costs can not be estimated at
the same accuracy of operating and investment costs. Increasingly regulators are
under pressure to quantify these costs and provide for their recovery in the price.

The literature on utility price regulation appears to have evolved separately

to that focused only on Pareto efficient pricing1 ; and tends to treat the regulatory
environment as imposing a degree of incentive on a firm to manage costs and
profitability. There are several forms of price regulation that we will review in the

following sections.

2.3.1 Cost of Service Regulation

Direct price and rate of return regulation are collectively termed cost of service
regulation. Direct price regulation affords the least incentive to the firm to operate
efficiently, and has been the most criticised on economic grounds (Harris, Tate,
and Renzetti, 2002).

Rate of Return price regulation allows the supplier to earn an agreed rate of
return on its capital assets. Assuming a single period, and if fixed overhead costs
are covered equally by fixed revenues (this might be achieved by a fixed charge
applied to each account holder during each billing period), then we only have to

consider the variable components of revenue and cost:

px = C(x, w) + rK

where p is the unit price, x is the unit consumption, r is the agreed rate of
1
Studies that tackle these issues together are uncommon.

13
return on K units of capital. C() is a variable cost function that depends on
consumption and a vector of input prices w. The price is therefore:

K
p = AV C + r
x

The capital output ratio K/x can be assumed constant for constant technology,

and so the price is simply the average variable cost plus a constant markup. When
the markup is implicitly included in the variable cost, the unit price is simply the
average variable cost. This corresponds to a Ramsey number of one.
In practice the required rate of return is decided by the utility and built into

its costs as presented to the regulator. The regulator sets the price such that
the present value of future cash flows is equal to the present value of historical
cash flows. The only decision variable is then the discount rate (often called the
appropriate discount rate - ADR) used to discount future cash flows.

Rate of return regulation gives a degree of comfort to investors because of

the guaranteed return and low risk. This can lower the cost of capital for the
owners of the utility and ultimately benefit consumers. On the other hand there
are a number of problems associated with this form of pricing. Rate of return

regulation limits the firms incentive to reduce costs because allowable costs are
always recoverable. Firms have been known to engage in cost padding in their
submissions, and deciding on allowable costs is problematic. There is also potential
for regulatory capture (Laffont and Tirole, 1991). Firms regulated under rate
of return tend to invest in excess of the efficient level if their cost of capital is

lower than the rate of return - the Averch-Johnson Effect (Averch and Johnson,
1962). Rate of return regulation has also been criticised for the monitoring and
reporting burden it places on both regulator and utility. Where different firms
service different areas, there may also be significant discrepancies in prices.

14
2.3.2 Price Cap Regulation

Price cap regulation (Littlechild, 1988) is mainly intended to increase cost

efficiency and promote investment. The firm manages its cost and pricing and
is permitted to keep its profits for the period that the cap is in place. The cost

reporting requirements are less; although the determination of the initial price cap
is reliant on firms historical and projected costs. The need to determine allowable
costs and the rate of return is removed, as is the need for monitoring the firms
profits.

Price cap regulation is prevalent in the telecoms and electricity markets of

Australia, where it operates under a form of allowable price adjustment known as
CPI-X. In the UK it is called RPI+X and covers the telecoms, gas, airports, water,
and electricity sectors. Under this regime, prices follow the CPI plus or minus a

constant rate (X) that is determined by the regulator. The value of X is set so
as to reflect cost efficiencies and technological improvements that are expected to
occur over time, promoting efficiency and investment by the firm. In the next
period, the base price is progressively lowered so that these benefits are passed
on to consumers without diminishing the profit incentive for the firm. Successful

implementation of price cap schemes depends on correctly setting the initial price
base and the level of X. In some instances incorrect settings of these values may
lead to problems. A price cap that is set too low may discourage investment;
it may be uneconomic for the firm to employ all its capital, creating a problem

of stranded assets and placing the regulatory agreement under strain (Newbery,
1998). On the other hand, firms that become too profitable may face a windfall
profits tax.
Price caps appear to work well when competing firms offer different services

using a monopoly network as is the case with the Australian mobile phone network.

15
The cap is set to an average for a basket of services and the firms are free to set
prices within the basket as long as the average price is below the cap. Recently

the NSW Independent Pricing and Regulatory Tribunal (IPART) has canvased the
idea that the monopoly Sydney metropolitan supplier, Sydney Water, should allow
third party access to the trunk network to supply competitive services (Keating,
2006).

2.3.3 Performance and Yardstick Based Approaches

Other forms of regulation including performance and yardstick regulation, have

a light handed regulatory approach. Instead, targets are set - normally based
on observation of firms regarded as industry leaders. These may relate to cost
efficiency; but may also focus on quality, management and staffing, or some other

measure of performance. In practice a hybrid form of regulation may be adopted.

These methods place more emphasis on statistical and econometric approaches
to benchmark setting. This is in contrast to cost of service and price caps which
rely on historical and future cost estimates, and use averaging or present value
methods to aggregate. Two methods that have received attention are Stochastic

Frontier Analysis (SFA) and Data Envelopment Analysis (DEA). Chapter 11 will
present the motivation for using and theory behind stochastic frontier analysis,
while Chapter 12 will give a case study application of this method.

2.3.4 Incentive Regulation Theory

The foregoing discussion indicates that regulation imparts some degree of in-

centive to a monopoly supplier to act as if it faces a competitive market, while

the regulator is faced with imperfect information about the suppliers operation.
Work in incentive theory views the regulator as facing an optimisation problem
that requires maximisation of expected social welfare subject to a constraint of

16
asymmetrical information. This situation involves a welfare loss compared to one
of perfect information as, for example, when the firm engages in cost padding.

The extent to which a regulator observes utilities true costs has been the sub-
ject of a number of studies. Baron and Myerson (1982) assumed that the regulator
is unable to observe costs and therefore the firm has no need for overstatement.
Laffont and Tirole (1986) present a model in which the regulator observes noisy

cost and output of the firm. The firm can adjust its efficiency (proxied by marginal
cost) according to an effort parameter; neither of which are observed by the reg-
ulator.
The problem of asymmetric information has been the subject of work related

to cost of service regulation of water utilities. In Wolak (1994) a procedure for esti-
mating the parameters of a regulated firms production function under asymmetric
information is presented based on a sample of California Class A water utilities.
The model assumes that the firm holds private information that it can exploit to
increase profit while revealing only the minimum requirement of information to

the regulator. In a recent paper, Bougheas and Worrall (2006) model cost padding
by observing differences in real and total (padded) costs in the contracting and
the post-contractual stages.
The standard incentive regulation model of Laffont and Tirole (1986) is a

Bayesian optimising problem which there is a trade-off between cost reduction

and rent transfer to consumers:

R = bR + (1 b)C

where b is the power of the regulatory incentive scheme, R is a base revenue

and C are costs. The lowest regulatory power, (b = 0), corresponding to rate of
return pricing, creates an opportunity for the firm to engage in cost padding. At

17
the highest power, (b = 1), the utility receives all the benefit of its cost reduction
and does not need to inflate its costs.

2.4 Practical Approaches to Urban Water Pricing

The practical basis of water pricing in a regulated environment is primarily

cost recovery based and reliant on the use of historical and forecast cost and

consumption data. The requirement for economic efficiency in water supply and
criticism of rate of return regulatory approaches has focused attention on marginal
cost approaches in price determination. Mann (1993) discusses the advantages of
marginal cost pricing in an American context. In this section we consider the

implications of a strict marginal cost approach before reviewing three different

industry approaches to urban water pricing: Turvey long run marginal cost, aver-
age incremental cost, and base extra method. Each of these are departures from
strict marginal cost but retain some of the desirable efficiency characteristics of
that approach.

Strict marginal cost pricing of water entails the use of two marginal cost
regimes (Warford, 2003). The price of water is its short run marginal operat-
ing cost as long as capacity is in excess of demand, and distribution requirements
are met. Marginal capacity cost is zero when existing supply is at a safe level

in excess of short term demand. The applicable marginal costs occur only in
operating and distribution. Once demand increases to the level where capacity
constraints are evident, prices increase (short run marginal cost becomes vertical)
until either demand is reduced or consumers reveal their willingness to pay for

the required investment in new capacity. If the latter is the case then the price
follows the long run marginal cost curve. Once the capital investment has been
carried out, there is again excess capacity and the relevant price is again the short
run marginal cost curve. Expansion of the distribution network can be treated

18
similarly, with marginal costs becoming zero once expansion is completed.
A variant of strict marginal cost pricing is based on the observation that de-

mand varies seasonally and therefore capacity constraints are more likely to be
evident at these times. In a study by Renzetti (1992), the welfare effects of adop-
tion of a peak load pricing scheme were analysed and found to be positive. This
entailed pricing according to the short run marginal cost in off peak demand pe-

riods and the long run marginal cost in peak demand periods (typically summer
months). Peak demand prices were also proposed to be increased further in the
event that capacity constraints would be met. Grafton and Kompas (2006) pro-
pose a modification to peak pricing that increases prices according to dam storage

levels - implicitly weighting the marginal supply cost by a scarcity factor. Again,
this would be subject to ensuring that the predicted aggregate demand at the new
price level was safely within existing capacity. When dams were at or close to full
storage levels, prices would revert to their short run marginal cost.
One feature of marginal cost approach to water pricing is the inclusion of a

fixed charge or connection fee. This is justified for two reasons. First, as a means of
recovery of fixed costs such as administration, meter reading, and other overheads.
Second, as a form of Ramsey pricing, to recover the variable cost losses that arise
when marginal operating costs are below average costs. Sibly (2006) argues that

equity considerations in the use of a fixed charge can be overcome by relating it

to property values - in the sense that these are a proxy for the consumers ability
to pay, or by use of a discount or concessional scheme based on need.
Although theoretically desirable, in practice strict marginal cost pricing is dif-

ficult to achieve for a number of reasons. Firstly it would lead to price fluctuations
that may be unpopular with consumers and hold political risks for government.
Secondly funds for investment in additional capacity may become available at a
time when a comfortable buffer of capacity is still available; causing consumers to

19
question the need for price increases. Thirdly there are risks involved with man-
aging supply at or near its demand level. Unforeseen pressures on the supply such

as increases in domestic demand, firefighting, or drought could lead to a critical

water shortage. The approach prefered by utilities and governments is to leave a
safe buffer of capacity.
To overcome these problems the utility industry has adopted several different

approaches which we outline in the next three sections. Primarily these aim to
recover costs, allowing for the lumpy nature of capital investment in water supply
and the desirability of non-volatile rates.

2.4.1 Turvey Long Run Marginal Cost

Early work by Turvey (1976) set out to define the long run marginal cost of

water supply as the sum of marginal capacity costs, marginal operating costs,
marginal distribution costs and per connection overhead costs. Turvey claimed
that marginal capacity costs need to be considered in a different light to conven-
tional economic theory. Because commitments to capacity expansion are made
years in advance to their implementation, the decision variable is not the actual

incremental system expenditure, but rather the timing of implementation. Turvey

shows that a one year change in commissioning new capacity can have a signif-
icant marginal cost, based on different capital recovery factors (discount rates)
and consumption. This, he argued, is the actual marginal capacity cost. The UK

Office of Water (OFWAT) (Price, 1993) and the Essential Services Commission
in Victoria (Victorian State Government, 2005) have both employed this concept
of marginal capacity cost to estimate long run marginal costs.

20
2.4.2 Average Incremental Cost Pricing

A second approach common among North American water suppliers is the

Average Incremental Cost (AIC). This is defined as the additional cost per unit of
additional consumption when both additional annual costs and consumption are

expressed as the present value of their flows (Warford, 2003). If price is equated
to cost of supply, the applicable volumetric price at the start of the period is:

PV (It + Ct C0 )
p = AIC =
PV (xt x0 )

where for year t, It is investment, Ct is operating cost, and xt is consumption,

and the base year is t = 0.
Whilst theoretically less attractive than strict marginal cost approaches, the

AIC approach has the desirable property of producing a constant price over for the
period (possibly adjusted annually for inflation). The difficulties of this approach
for the regulator are that the discount rate remains to be determined, and there
is a possibility of a price discontinuity arising from a period in which no major

investments are undertaken.

2.4.3 Base Extra Method

The Base Extra Method, used by the American Water Works Association
(AWWA), is an allocation of historical costs and demands by user class and cost
category. The user classes are residential, industrial, and commercial; and the cost
categories are base, extra capacity, customer service, and fire protection. Base and

extra capacity refer to marginal costs of production and investment under normal
and peak consumption, while customer service are billing and metering overheads.
Fire protection is related to cost of hydrants. A formula is used to determine the
ratio of base to extra capacity costs. Each cost category for each user class is

21
assigned a rate method class cost which is the average unit cost of providing that
service to that user class. For example, the rate method class cost of base demand

for the residential class is the average cost of one cubic meter of demand under
base demand. The rate method class costs are summed to determine the tariff
rate for each class of user.
There are two points to note in respect of the Base Extra Method. Firstly, it

results in unit prices that are higher for residential users (compared to industrial
and commercial) as that class are more likely to be assigned extra capacity costs.
Secondly the price schedules will need modification if a block rate tariff is required
or if the regulator disallows certain costs.

2.5 A Functional Approach to Water Pricing

The foregoing has demonstrated that regulation of utility industries requires

the quantification of many variables including tariff price, input factor prices,
allowable costs, social and private discount rates, industry rates of return, per-

formance and quality standards, social and environmental costs and externalities,
and mechanisms for their recovery. The utility must be able to predict population
growth to determine capital investment requirements for the rate setting period.
These capital investments must be ready well in advance of the time that predicted

increases in demand occur. The sources and timing of finance are important fac-
tors also. These may not be directly controlled by the utility, and therefore short
term adjustments in the form of price increases or restrictions may be required
to cope with delays in investment. The regulator must ensure the presence of

equity in pricing - so that low income groups receive a basic quantity of water
at an affordable price. The regulator must also ensure that the supplier remains
financially viable and is able to recover costs and repay creditors.
From an economic perspective the problem of water pricing requires an under-

22
standing of the opportunity costs of production, and, in the absence of competitive
markets, determining how costs and benefits are allocated among consumers, pro-

ducers and the environment. Current industry approaches have several deficiencies
from an economic viewpoint. Firstly, there is an emphasis on average cost instead
of marginal cost approaches to water pricing. While the average cost approach
means that costs are recovered, it is also a less efficient solution as price signals

are incorrect. This is particularly important when scarcity needs to be considered.

Secondly, there is a dependence on historical costs in price determination. From
an economic perspective these are sunk costs and do not enter into the decision
mix because they cannot be controlled. Thirdly, existing methods based on sec-

ond best solutions, may result in higher prices for users whose demand is inelastic
because they have no substitutes.
One could argue that a market based approach to urban water supply may
resolve some of these problems. Indeed, Littlechild (1988) sees regulation as a stop
gap measure until competitive markets are in place. It is difficult to see how this

might become completely possible given that most urban water is delivered over
a network. Instead, competition, when it is introduced is likely to appear in other
areas that can be dissociated from the network itself, such as meter reading and
billing, maintenance, competitive tendering for capital works, and public-private

partnerships. Importantly, as we will see in Chapters 11 and 12, the increased use
of performance based benchmarking means that competition will not be within
a network area but across different networks - municipalities, regions and indeed
countries.

In the chapters that follow we depart from current industry and regulatory
approaches to water pricing to examine the problem from a purely functional
stance. We develop cost and demand functional specifications and, in case studies
that follow, we test these against a number of data sets. This will allow us to

23
characterise the parameters of production, cost and demand such as economies of
scale and elasticities; and to demonstrate the utility of this approach to pricing.

24
 CHAPTER 3

Production and Costs

3.1 Introduction

The objective of this chapter is to introduce production and cost models suit-

able for analysis of urban water service providers. Apart from their use in de-
termining the marginal cost function, these models will enable us to measure the
economic parameters of production and cost including output elasticities, cost
shares, returns to scale and economies of scale. This introduction contains an
overview of some of the work that has been carried out in this area, with em-

phasis on the various models employed. This is followed by a presentation of the

theoretical basis of production and cost models. We then develop a minimum
cost model for both constant returns to scale production technology and one for
variable returns to scale production. This is followed by an examination of issues

in estimation and the transformations required to express the models in a form

suitable for estimation. Actual estimation of cost functions will be presented in
two case studies that commence in Chapter 6. We conclude this chapter with a
brief discussion.

3.2 Production and Costs in Urban Water Supply

An urban water supply can be broken down into several components. The
raw water is extracted from a source such as a reservoir, aquifer, river or the sea.

25
It is then treated in various ways including filtration, chemically treated to meet
public health standards for potable water, possibly treated for taste, and in some

countries fluoridated. The water is distributed via a network of pipes to consumers

including residential, commercial, and industrial users. For the purposes of this
study, the extent of this network defines the boundaries of urban - in contrast to
a rural supply which services outlying communities and relies on other forms of

distribution such as channels, or where consumers source their own water.

Wastewater is collected via the sewerage network to treatment plants where
it is treated to a standard suitable for discharge to the sea or some other water
source. Stormwater is runoff collected by sewers or a separate drainage network

and enters the wastewater treatment system or is dispersed directly to the sea
or other water source. Leakages occur throughout water networks which result
in unaccounted for water. Wastewater and stormwater networks may also record
losses. Network infiltration is not uncommon also, and this can be a major problem
if the potable supply is infiltrated.

Production is generally measured in terms of the amount of water delivered

to customers, that is, the output of treatment plants. Generally this is not the
same as the amount of water extracted from its sources as losses occur in the
extraction and production. Similarly, wastewater volumes do not generally match

consumption because of losses. Costs are the costs of management, operating,

and investment in the various components of supply. These include the facilities
where the water is harvested, treatment plants, pumping facilities, distribution
networks, storage facilities, service connections, meters and fire hydrants. For

wastewater the components include the distribution network, treatment plants,

and may also include facilities for solid waste/effluent disposal. Costs can be
classified as centralised or distribution costs. The distinguishing feature is that
centralised costs are shared by a greater number of users compared to distribution

26
costs which are incurred for fewer users as network density decreases. This explains
why marginal costs are greater as network density decreases, and also why costs

for network expansion, in for example, residential developments, are often born
by the users.
The environment contributes to and is impacted by urban water supply. River
basins act as collectors for reservoirs, and pollution of rivers by agricultural, in-

dustrial, or human activity can contribute to treatment costs. For supplies that
rely on groundwater, poor quality and over extraction can also have an impact
on treatment costs. Urban water and wastewater treatment are both energy in-
tensive and therefore contribute to greenhouse emissions. Wastewater treatment

itself is an industrial process that requires large amounts of water inputs. Waste-
water is returned to the sea or river systems, while solid waste residues go to land
fill. Environmental costs are difficult to quantify and applied work in this area is
uncommon. In a number of countries water authorities now include levies in con-
sumer tariffs which are returned to government to fund environmental initiatives.

3.3 Functional Modelling of Production and Cost

The estimation of production and cost functions of particular industries or

sectors either nationally or regionally holds considerable interest for researchers.

Work undertaken in this area has focused on the impacts of model specification,
elasticity of substitution, and the dynamics of capital adjustment. Based on a US
manufacturing data set, Kim (1992) used a translog production function to show
that assumptions about homotheticity, homogeneity, and constant returns to scale

in production were incorrect. In a study of industrial factor demand across six

OECD countries, Kolstad and Lee (1993) also used a translog cost functional form
to examine the variability of results under different model dynamics. The models
specified include long run cost (full static equilibrium), short run restricted cost

27
(partial static equilibrium), and short run restricted cost with a Euler equation
(partial dynamic equilibrium) governing the change in capital stock. In a similar

study, Bregman, Fuss, and Regev (1995) estimated production and cost functions
for a panel data set of Israeli industry for the period 1979 to 1983. In each case
short and long run estimates were obtained under assumptions of cost minimising
choice of inputs. In Balistreri, McDaniel, and Wong (2003) the authors estimate

the short and long run elasticity of substitution for different industry sectors
based on U.S. Bureau of Economic Analysis (BEA) data and suggest that the
true relationship between capital and labour is indeed most likely to be Cobb-
Douglas. The study employed time series aggregates of production and inputs

and compared results based on AR(1), first differences, and single error correction
models.
For studies that are focused on the provision of utility services, including water,
major areas of interest are model specification, long run marginal cost, returns
to scale, and application of Ramsey pricing principles. We reviewed the major

developments in this area in the previous chapter. As was discussed, one feature
of current water pricing practice is the use of discounting to aggregate historical
cost and future cost estimates - in contrast to the use of statistical modelling of
cost functions. This approach is characterised by the two main approaches to cost

determination used in the UK, USA and Australia: the Turvey method and the
Average Incremental Cost method.
A functional approach to cost modelling has been a feature of the research
literature since work by Nerlove (1963), considered to be among the earliest ma-

jor application of statistical cost analysis (Greene, 2003, p. 125). Hines (1969)
studied the relationship between cost and utility size in water production by re-
gressing average fixed, variable and total costs on plant capacity utilisation and
adjusted plant investment. These latter two terms are engineering measurements

28
of utilisation and capital inputs. Bruggink (1982), in the context of the debate
concerning public vs private ownership, modelled the effect of ownership on techni-

cal efficiency of municipal water companies in the USA. He used a linear model to
regress operating costs on production and a set of variables related to inputs and
input costs, treatment methods, distribution, regulation, and ownership. Feigen-
baum and Teeples (1983), also contributing to the ownership debate, criticised the

use of single output models and instead applied a hedonic model in translog form
that included a multiple output variable containing volume produced and a vector
of service attributes. Bhattacharyya, Parker, and Raffiee (1994) used a translog
form for the cost equation to estimate technical, price, and scale efficiencies based

on a cross section of American water utilities to conclude that public utilities were
on average more efficient. Wolak (1994) uses a Cobb-Douglas production form to
derive a two factor variable cost function in a study of regulator-utility informa-
tion asymmetry. More recently, Saal and Parker (2004) examine the impact of
privatisation of regional water authorities in England and Wales by measuring

changes in productivity growth using a time dependent translog cost function.

There appear to be relatively few studies that directly address issues of welfare
and tariff design. In a departure from other cost studies, Renzetti (1999) exam-
ines the welfare loss brought about by consumption at current prices compared to

consumption at marginal cost for a cross section of municipal water and waste-
water utilities in Ontario, Canada. He firstly estimated a system of cost and cost
share equations for water and wastewater using a translog cost specification, and
then estimated aggregate demand for residential and non-residential consumers

using an OLS specification. The study found evidence of increasing returns to

scale in both water and wastewater operations. Welfare loss due to mispricing
was found to be around 78% of the existing unit price and around 29% of the ef-
ficient marginal cost for residential consumers. In another study (Pushpangadan

29
and Murugan, 1998), the authors aim to address the problem of equity in Ramsey
pricing of water supply under increasing returns to scale. They estimate a partial

equilibrium to determine an optimal tariff for Kerala State in India that allows
for subsidisation of low income (inelastic demand) users by high income (elastic
demand) users.

3.4 Theory of the Firm

3.4.1 The Constant Elasticity of Substitution Production Function

The purpose of this section is to review the major results related to the theory
of the firm. We will initially restrict ourselves to a class of production functions

having those properties that permit analytical solution of the profit maximisation
problem:

homogeneous of degree one in factor inputs (ie. production exhibits constant

returns to scale)

quasiconcave

twice differentiable

Within this class of functions, the Constant Elasticity of Substitution (CES)

(Layard and Walters, 1987, p. 272) production function exhibits the required

properties and has been widely used in economic analysis. The CES production
function expresses a firms output x as a function of its inputs, capital K and
labour L:

1/
x = K + (1 )L


where is an efficiency parameter, a substitutional parameter, and is a

distributional parameter.

30
 The implicit form of the CES function is:

F (x, K, L) = x K (1 )L = 0

By the Implicit Function Theorem (Klein, 1998, p. 240) we get the marginal
products of labour and capital:

FL  x 1+
xL = = (1 )
Fx L

FK
 x 1+
xK = =
Fx K

The ratio of these two is:

 1+
xL (1 ) K
= (3.1)
xK L

The elasticity of substitution measures the change in the input ratio as the

ratio of marginal products changes:

log(K/L) 1
SKL = =
log(xL /xK ) 1+

Production technology is determined by the limiting value of . When this is ,

the technology is Marx-Leontief, when it is 1, it is linear or perfect substitution.
As 0, SKL 1, and the technology is Cobb-Douglas production.
Totally differentiating the implicit CES form:

()x1 dx ()K 1 dK (1 )()L1 dL = 0

we set 0 and integrate:

31
 dx dK dL
= + (1 )
x K L
log x = + log K + (1 )log L

to obtain the Cobb-Douglas production function:

x = K L(1) (3.2)

This function satisfies all our required properties. Moreover, the homogene-
ity requirement means that this class of production technology exhibits constant
returns to scale, a linear increase in both inputs results in the same increase in
output.

Under Cobb-Douglas, equation (1.1) above becomes

 
xL (1 ) K
=
xK L

This equation reveals two important properties of Cobb-Douglas technology.

Firstly, the slope of all isoquants are the same for constant capital labour ratio.
Secondly, moving along an isoquant (increasing capital labour ratio) the ratio
of marginal product of labour to marginal product of capital is increasing. The
same level of output is achieved with less labour so its marginal product must

be increasing. By the same argument, the marginal product of capital must be

decreasing.

3.4.2 The Constant Returns to Scale Minimum Cost Function

In the following treatment we assume that in water supply, input and output
prices are determined exogenously and the firm has decided on a short term level

32
of production output. Therefore the firm can only adjust input levels and will
seek to maximise profits by cost minimisation. To simplify notation, we adopt a

Cobb-Douglas production function with all inputs variable.

The firms cost minimisation problem is to minimise:

C(w, r, L, K) = wL + rK

subject to the Cobb-Douglas production constraint:

x = K L(1)

where w, r are the factor prices. The first order conditions are:

 
K
w = (1 )
L
 1
K
r =
L

The input price ratio is proportional to the input factor ratio, and also equal
to the ratio of marginal products:

  
w 1 K xL
= =
r L xK

Rearranging the first equality and solving for rK and wL in the cost equation,

we can see that determines the cost shares of each input:

rK = C

wL = (1 )C

33
 We express the production constraint in terms of the capital labour ratio and
replace this with the price ratio:

   
w
x = L
r 1

which can be rearranged to get the conditional labour demand function L(x, w, r).
Transforming the cost objective function:

  
K
C =L w+r
L

and inserting the expressions for labour demand and capital labour ratio de-
livers the minimum cost function for Cobb Douglas technology:

1 (1)
C(w, r, x) = (1 )1

r w x (3.3)

This function has the following properties (Luenberger, 1995, p. 42):

1. Homogeneous of degree one in factor prices

2. Homogeneous of degree one in output

3. Nondecreasing in factor prices

4. Concave in factor prices

3.4.3 Nonconstant Returns to Scale Minimum Cost Function

Despite the strong empirical evidence that production in many industries ex-
hibit constant returns to scale in the long run, there are reasons for relaxing this
assumption and seeking a more flexible technology. Utilities and other network
industries have barriers to entry that promote economies of scale and therefore

34
formation of natural monopolies (Luenberger, 1995, p. 64). Regulators of essen-
tial services require utilities to retain spare capacity. In the case of water supply

this is to ensure they can cope with peak demands and emergencies such as wa-
ter shortages or firefighting. This reserve capacity results in decreasing marginal
costs and therefore increasing returns to scale. In the short run, utilities may be
in a period of short term adjustment, they may be operating with an excess of

capacity that creates a decreasing marginal cost, or alternately, capacity may be

insufficient to meet demand, forcing marginal costs to increase.
In this section we present the derivation of the minimum cost function subject
to a production constraint in a similar manner to the CRS formulation above but

without the assumption about constant returns. We include three factors; capital,
labour, and energy. These models are sometimes termed KLE production models.
The firms cost minimisation problem is to minimise:

C(w, r, s, L, K, P ) = wL + rK + sP

subject to the Cobb-Douglas production constraint:

x = AL K P

where w, r, s are the factor prices. The sum of the exponents is the scale

economies parameter: Q = + + ; 0 1, 0 1, and 0 1. Note

that Q = 1 implies CRS technology and Q < 1 is decreasing returns to scale.
The first order conditions yield three price ratios:

35
 w K
=
r L
w P
=
s L
r P
=
s K

We use these to decompose the variable cost function into its shares, for ex-
ample:

    
K P
C = L w+r +s
L L
Q
= wL

The cost shares are: wL = Q
C, rK = Q
C, sP = Q
C.
We use the production function to derive the conditional factor demand for
labour:

"     # Q1
w w
L = A1 x
r s

Finally we substitute this expression into the labour cost share to get the
unrestricted cost function:

"     # Q1
w w w
C=Q A1 x
r s

  Q1     Q   Q
1 Q 1
C(w, r, s, x) = Q (x) Q (3.4)
A w r s

This function has the following properties, which are verified in Appendix A:

36
 1. Homogeneous of degree one in factor prices

2. Homogeneous of degree (1/Q) in output

3. Nondecreasing in factor prices

4. Nondecreasing in output

5. Concave in factor prices

3.5 The Translog Cost Functional Form

The preceding two cost models are either constant in output or exponential
functions that are decreasing or increasing in output. In each case economies

of scale are independent of factor prices. The function will indicate increasing,
constant or decreasing returns to scale over the full range of output. In the case
that economies of scale changes with output from increasing to decreasing returns,
for example because average cost is quadratic in output (the cost function is a
cubic polynomial), or that cost factor price elasticity is not constant, the CES

functional forms may be too restrictive.

There are several approaches to this. The first is the direct use of a linear cost
model that includes higher order terms as regressors. Although text books often
use this form to show the parabolic shape of cost functions, in empirical work, this

approach is not popular because the analytic form of cost elasticity also involves
quadratic terms. The second is to employ a log-linear form that includes a squared
log output term. Cost elasticity is a linear function of log x, but there is still the
question as to how prices enter the quadratic model.

A third approach involves the use of a more flexible cost function. The translog
(transcendental logarithmic) function is widely used in the estimation of produc-
tion and cost functions in empirical economics. It can be used to estimate any

37
arbitrary polynomial function, however in practice only second or third order
functions are used.

A translog function is a Taylor series expansion about some value of its argu-
ments. As an example, Bhattacharyya, Parker, and Raffiee (1994) used a translog
cost function to estimate the parameters of a variable cost function for a sample
of US water utilities. The authors used a function with one input and output

variable and three price variables Z = (X, p), |Z| = k. The second order Taylor
series was expanded around log(Z) = 0 :

k k k
X 1 XX
log C = i log Zi + ij log Zi log Zj
i=1
2 i=1 j=1

Translog cost functions are usually estimated as a system of equations. We

will discuss issues related to their estimation in the next section.

3.6 Estimation of the Production and Cost Functions

3.6.1 Issues in Estimation

Before estimating these models there are several issues that we need to con-
sider.

Firstly, the theoretical models based on CES production are static models that
implicitly assume use of a cross-sectional data set. If multiple observations over
time are available we would like to make use of these. The use of time series
data introduces problems of autocorrelation of the error term and nonstationarity
of time series. Autocorrelation can be tested for by use of the Durbin-Watson

statistic and estimation can proceed using a technique such as Cochrane-Orcutt

estimation (Pindyck and Rubinfeld, 1991, p. 141). We can test for nonstationarity
of time series using Dickey Fuller tests. Estimation in the presence of nonstationar-
ity of time series can proceed if the series for the dependent variable is cointegrated

38
with a linear combination of the series for the independent variables. Otherwise
an OLS model runs a risk of spurious regression and the model specification must

be altered.
Secondly, a problem often encountered in applied work is omitted variables
bias. The occurs when an unobserved independent variable is correlated with other
independent variables. Varian (1984, p. 175) gives an example of this in relation

to agricultural production. Suppose there is a factor land quality that is related

to both output and to choice of capital inputs. If this is unobserved it becomes
part of the error term. However under the OLS assumptions the error term must
not be correlated with the regressors and so estimates will be biased. Wooldridge

(2001, p. 248) mentions three methods for overcoming omitted variables bias in
cross sectional data: OLS with a proxy for the omitted variable; instrumental
variables method such as 2SLS; and a multiple indicator instrumental variables
procedure.
Omitted variables is one instance of a more general problem of specification

error. The generally accepted approach to determining the presence of specifi-

cation error is by use of various tests, for example the Regression Specification
Error Test (RESET) test (Hill, Griffiths, and Judge, 2001, p. 187). The use of
panel data, if available, is another method for overcoming the omitted variables

problem.
Panel data studies can be characterised by the presence of some heterogeneity
among cross-sectional units that does not alter over time but that may not be
directly measurable. This heterogeneity or unobserved effect causes bias in OLS

estimation if it is correlated with any of the explanatory variables, including lags

of those variables. With a two period panel data set, for example, a time-constant
unobservable variable can be eliminated by use of first differencing. If the OLS
conditions of orthogonality and full rank based on difference variables still hold

39
then OLS will produce unbiased results. A key requirement for maintenance of
the orthogonality condition is strict exogeneity. This is the requirement that the

regressors in any time period are not correlated with the error term in any other
time period. That is, unobserved effects that are absorbed into the error term
may not influence explanatory variables in another time period.
For utility industries, estimation of the minimum cost function and use of du-

ality to derive the parameters of the production function has an important advan-
tage over the direct use of the production function. If factor prices are exogenous
then any omitted endogenous factor (unobserved effect) will be uncorrelated with
prices (Varian, 1984, p. 178). In the cost model presented earlier, output is also

an explanatory variable. It is reasonable to assume that water output is driven

by exogenous demand and so will be uncorrelated with the error term.
A third issue that has hampered progress in empirical work is difficulty in the
measurement of capital stock. Most water infrastructure assets in the major cities
of the world were commissioned many decades ago. Accounting book values may

differ from replacement value. The outstanding debt on assets might be used as
a capital stock measure but this also may not reflect replacement value. This
provides further motivation for use of duality in estimation.

3.6.2 Specification of the Constant Returns to Scale Minimum Cost Function

The Cobb-Douglas production function is:

x = K L(1)

By confining ourselves to a restricted model, and with cross sectional data, a

log-linear form may be expressed in terms of a single regressor:

40
    
xi Ki
log = 0 + 1 log + ui
Li Li

where ui is an independently and identically distributed error term with zero

mean.
The minimum cost function corresponding to this production function is:

C = Ar w 1 x

In restricted CRS log linear model form with cross sectional data this becomes:

   
Ci ri
log = 0 + log + ui
wi xi wi

Note that the parameters of the production function based on estimation of

the minimum cost function will in general not be the same as those derived by
direct estimation.

3.6.3 Specification of the Variable Returns to Scale Minimum Cost Function

The three factor unrestricted minimum cost function is:

  Q1     Q   Q
1 Q 1
C=Q (x) Q
A w r s

The log linear form requires factoring out the coefficients and forming price

ratios. Using cross sectional data:

     
Ci wi si 1
log = 0 + log + log + log (xi ) + ui
ri Q ri Q ri Q

In this form the variable coefficients form a system of three equations with
three unknowns and hence a unique solution for the estimates exists. The choice

41
of which variable to form the ratio is arbitrary.

3.6.4 Translog Cost Function

We assume cost is a product of two separable output and price functions, then:

log C(X, p) = log f (X) + log g(p)

In the case that there are constant returns to scale, then g(p) is the average

cost function.
Shephards lemma states that if C(X, p) is a minimum cost function, then the
cost shares that produce the minimum cost for some level of output are:

log C(X, p) log g(p)

si = =
log pi log pi

The second order Taylor series expansion of log g(p) about log p = 0 is:

k k k 
2 log g

X log g 1 XX
log g = 0 + log pi + log pi log pj
i=1
log p i 2 i=1 j=1
log p i log p j

The derivatives in this expression become constants when evaluated at the

expansion point. Let

log g 2 log g
i = , and ij = ji =
log pi log pi log pj

then we have:

1 1
log g = 0 +1 log p1 + +k pk +11 log2 p1 + +kk log2 pk +12 log p1 log p2 +. . .
2 2

We derive the ith share equation by differentiating this with respect to the ith

42
price:

log g(p)
si = = i + i1 log p1 + i2 log p2 + + ik log pk
log pi

In the share equation form, the elasticities of substitution (Greene, 2003,

p. 368) are ij = (ij + si sj )/(sisj ) and ii = (ii + si si si )/(s2i ) .

The share equations form a system of k equations that can be estimated by

a multiple equation method such as Seemingly Unrelated Regression (SUR) esti-
mation (Greene, 2003, p. 366).

3.6.5 Panel Data Estimation

The basic linear regression model used in panel data estimation is:

yit = xit + zi + it

where xit is a vector of explanatory variables for the ith cross sectional unit, and

zi is a vector of time constant individual effects for the ith cross sectional unit. An
example (Greene, 2003, p.284) of the presence of time constant individual effects
in short and wide panel data analysis occurs when different firms exhibit different
economies of scale.

We consider three different techniques that may be used to estimate the pa-
rameters of the equation above depending on the form of zi .

1. Pooled regression. When zi is a constant then OLS will yield consistent and
efficient estimates of the parameters and the constant term .

2. Fixed effects. This entails capturing the mean group specific behaviour in

a single variable that is constant for each cross sectional unit. This means
that we are replacing the unobserved linear combination of individual effects

43
 with a single value, ie. i = zi .

3. Random effects. In the random effects approach we split the individual

effect into a term that is constant mean value for the whole panel (as in
the pooled case) and a disturbance term distributed around this mean but
with only one random drawing for each cross sectional unit. The regression
model appears now as:

yit = xit + + ui + it

with the constant term capturing mean behaviour for all units and ui the

time constant cross sectional disturbance term.

3.7 Chapter Discussion

One of the objectives of this thesis is to establish an economic basis for urban
water pricing based on marginal cost. This chapter has presented several pro-
duction and cost models that appear promising for this application. Much of the
applied literature has focused on the use of cost models in applications other than
price determination. These include measurement of economies of scale and scope,

elasticity and substitution effects, and cost drivers. Duality can be used to derive
the parameters of production and factor demands. Cost modelling is important
for identification of endogenous and exogenous cost drivers. Shadow pricing can
be used to price non-market factors, in particular those that are associated with

user costs. The theory underlying these cost models assumes that firms aim to
minimise costs. In the case of water utilities, this is reasonable given that there
are few other controllable parameters. Later in the thesis we will contrast this
with another approach that relaxes this assumption.

44
 CHAPTER 4

Household Demand

4.1 Introduction

In this chapter we present the theoretical basis for determination of the urban

water demand function. Determination of price and income elasticity of demand is

generally the primary aim of empirical demand analysis (Deaton and Muellbauer,
1983, p. 17). For the purposes of this study, an estimate of household demand will
help determine the efficient price, where net benefits are maximised. For water
authorities, it is important to know how households will respond to changes in

prices. In the short run, demand can be expected to decrease with an increase in
price. The change in revenue may be positive or negative depending on demand
elasticity. For a scarce resource such as water, these changes need to be balanced
with the benefits of reduced demand. Changes in demand can also be expected as

incomes increase. A high income family may eat out more often, and reduce their
domestic water consumption, compared to a lower income family. Alternatively,
they may build a bigger house with more bathrooms and consume more water.
We present two complementary models, both have gained considerable accep-

tance in empirical work over more than two decades. The first is the Almost Ideal
Demand System of Deaton and Muellbauer (1980) that builds on early work in
linear expenditure systems. The second is a model of household demand in the
context of block rate pricing - a form of pricing common in utility industries. This

45
model was proposed initially by Burtless and Hausman (1978). We describe esti-
mation of this model by use of a Maximum Likelihood Estimator (MLE), based

primarily on work by Moffitt (1986). This modelling and estimation framework

has been applied in a working paper from Rietveld, Rouwendal, and Zwart (1997)
based on a household consumption data set from Central Java in Indonesia. A
derivation of the MLE likelihood function used in estimation of the block rate

demand function is included in Appendix B.

4.2 Consumer Optimisation Problems

A consumer derives an amount of well-being or satisfaction through consump-

tion of a market basket, the cost of which does not exceed the available income
in the same period. This is measured by a utility function that permits different
baskets to be ranked numerically so that higher ranked baskets are preferable to
lower ranked ones. Utility, in its simplest form, can be expressed as a function of

consumption:

U = u(x)

where x is a consumption vector of goods in the market basket, or bundle.

Existence of a utility function is predicated on the axioms of choice: reflexivity,
completeness, transitivity, continuity, monotonicity, nonsatiation, and convexity1 .
These dictate a consistent ordering of preferences over bundles of goods (Varian,

1984, p. 113). In applied work, utility is usually not observable. If we assume the
existence of a utility function, we can use a constant unspecified level of utility,
represented by an indifference curve, to solve consumer optimisation problems in
terms of the variables that can be observed: prices, quantities and expenditures.
1
Formally, nonsatiation and convexity are not required for existence of a utility function.

46
 There are two constrained optimisation problems that are the basis of demand
theory. First, the utility maximisation problem is:

maximise u(x), subject to (p x) m

where p and x are vectors of prices and quantities for goods in the bundle,
and m is income. In this problem all points on the indifference curve that are
above the budget line are infeasible, those below the budget line are suboptimal

as there is a feasible indifference curve with a higher level of utility. The solution
to this problem is the indirect utility function v(p, m), the locus of maximal utility
obtained over the domain of prices and incomes.
Second, the expenditure minimisation problem is:

minimise (p x), subject to u(x) u

where u is the minimum required utility. In this problem all points on the
indifference curve that are above the budget line are suboptimal as they have
higher expenditure, those below the budget are infeasible, as they are below the
required level of utility. The solution to this problem is the expenditure function
e(p, u ), the locus of minimal expenditure (or cost) obtained over the domain of

prices and utility.

4.2.1 Demand Functions

Demand functions determine the response in the quantity of a good demanded

when prices, income, and utility are variable. The Marshallian demand curve
plots the change in consumption of the good as its price, or the price of any other

good changes - holding income at some constant level m . For any vector of prices
p, the utility maximisation problem produces an optimal level of consumption:

47
 xi (p, m ) = arg max u(x) , subject to (p x) m
xi

More generally we allow incomes to vary also, and the Marshallian demand
function defines a family of demand curves x(p, m). When attention is focused on
one particular good, and the consumers response to its own price, the remainder

of the market basket can be aggregated into one good (the numeraire) with price
normalised to one. The price of the good being studied is expressed as a relative
price and analysis is reduced to two dimensions.
The Hicksian or compensated demand curve, in contrast, plots the change in

consumption as prices change, for the expenditure minimising consumer - holding

utility constant. At each new price point, the consumer is compensated with more
or less income so that she remains at the same utility level u . Formally, Hicksian
demand is stated as:

hi (p, u ) = arg min e(p, u ) , subject to u(x) u

xi

More generally, allowing utility to vary also, the Hicksian demand function

defines a family of demand curves: h(p, u).

4.2.2 Applied Demand Analysis and the Almost Ideal Demand System

Indirect utility, Marshallian and Hicksian demand are functions that describe
consumer demand subject to different variables being held constant and under
different optimisation objectives2 . Applied analysis generally involves the search

for a functional form for demand that performs well in empirical work and satisfies
desirable theoretical properties. One such demand function is the Almost Ideal
Demand System of Deaton and Muellbauer (1980) and a modified version of it
2
These functions are related by a set of identities see: Varian (1984, p. 126).

48
that has proven popular in empirical work, the Linear Approximate Almost Ideal
Demand System. These are based on early work in linear expenditure systems,

for example, Pollack and Wales (1978).

The Almost Ideal expenditure function determines the least amount of money
required to reach utility level u when prices are p. It is a function of prices and
utility that is non-linear in its parameters, , , :

X 1 XX Y
ln e(p, u ) = 0 + k ln pk + kj ln pk ln pj + u 0 pk k (4.1)
k
2 k j k

In the following steps we eliminate the (assumed unobservable) utility u to

produce a form suitable for estimation. We begin by differentiating the expendi-
ture function with respect to the log price of good i:

ln e(p, u) 1X Y
= i + (ki + ik ) ln pk + u0 i pk k (4.2)
ln pi 2 k k

Shephards lemma (Jehle and Reny, 2001, p. 36) states that the Hicksian de-
mand for the ith good is equal to the change in expenditure for a unit change in
its price:

e(p, u)
= xi
pi

The left hand side of equation 4.2 is the price elasticity of expenditure. We
can show that this is equivalent to the share of expenditure for the ith good by

removing the logs and applying Shephards lemma:

ln e(p, u) e(p, u) pi xi pi
= =
ln pi pi e(p, u) e(p, u)

Now we express expenditure in terms of the desired level of utility u , and

49
assume that the minimum expenditure at this level is equal to income m. The
share equation can then be expressed in terms of prices and income:

xi pi 1X Y
= i + (ki + ik ) ln pk + u0 i pk k
e(p, u) 2 k k
1X
si = i + (ki + ik ) ln pk +
2
k
!
X 1 XX
i ln m 0 + k ln pk + kj ln pk ln pj
k
2 k j

The share equation is therefore:

1X m
si = i + (ki + ik ) ln pk + i ln (4.3)
2 k P

where P is a price index of the form:

X 1 XX
ln P = 0 + k ln pk + kj ln pk ln pj
k
2 k j

A simplified form of Equation 4.3 that has gained some popularity in empirical

work involves the use of Stones price index; a share weighted average price:

X
ln P = sk ln pk
k

The model that uses this price index is known as the Linear Approximate
Almost Ideal Demand System. The uncompensated demand elasticities (Green
and Alston, 1990) are:

ln P
 
1
ij = ij + ij i
wi ln pj

where ij = 1 for i = j otherwise ij = 0, and P is the chosen price index.

50
 The main advantage of the Linear Almost Ideal model appears to be data
related. Household expenditure surveys often record expenditure and prices, but

not consumption on different goods. Official price indices for commodity groups
can be used in preference to household surveyed prices. Use of this model in
applied work continues to be quite popular for example, Blanciforti and Green
(1983); Nelson (1994).

4.3 Modelling Demand subject to a Block Rate Tariff

Commonly, utilities such as water, power, and gas, are sold using a block rate
tariff, where the volumetric price varies with consumption, usually in non-linear

steps. Utility tariffs may also include a fixed service or connection charge, which
may contain a basic quota of the commodity, beyond which volumetric pricing
starts. Volumetric prices under a block rate tariff are either all increasing or all
decreasing. Decreasing block rates are commonly used when there are increasing
returns to scale and average costs are decreasing. Increasing block rate tariffs

(IBRT) are employed to reduce consumption when there are decreasing returns
to scale and average costs are increasing. When there are short run increasing
returns to scale, an increasing block rate tariff prices the commodity according to
its long run marginal cost so that investment in expanded capacity can occur.

In many parts of North America, domestic water was in the past priced using
a flat fee (decreasing average rate) or a declining block rate. This was a result of
the utility using average costs to price water and an abundant supply (Tietenberg,
1992, p. 241). Australia and the UK have progressively moved to volumetric

pricing with one or two block rates. As metering becomes more widespread, and
as the economic costs of water supply are better understood, use of IBRT pricing
is becoming more common.
The microeconomic theory of demand subject to a block rate tariff involves

51
representation of the budget constraint in a piecewise linear form that consists
of segments of constant marginal price and kinks where the price changes. Util-

ity maximisation subject to this kind of constraint induces a nonlinear form of

demand function that, as observed in empirical work, causes demand to cluster
around the kinks. There has been a considerable effort over more than thirty
years in theory development and model estimation in a variety of applications

where price is endogenous because it varies with the dependent variable. Early
studies commenced with the work of Burtless and Hausman (1978) in the con-
text of taxation and transfer programs, consumer surplus (Hausman, 1981), and
later in model specification and estimation by Moffitt (1986, 1990). Outside of

the labour supply and welfare literature, utility pricing using block rate tariffs is
an area of active research that has received much attention including Hewitt and
Hanemann (1995), Dalhuisen, Florax, de Groot, and Nijkamp (2003), Chicoince
and Ramamurthy (1986), Billings (1987), Taylor, McKean, and Young (2004),
Gaudin, Griffin, and Sickles (2001), and Nieswiadomy and Molina (1989).

4.3.1 The IBRT Demand Function

In the following we consider a household as the sample unit and confine our-
selves to a two good consumption basket where the second good is the numeraire,
normalised to have unit price. A linear budget constraint defines the set of feasible
consumption bundles whose cost does not exceed the available income3 . On the

constraint, income and expenditure are equal:

mi = pxi + yi

In this expression mi is household is income, p is the relative price of the first

3
We will assume that all income is spent.

52
good, xi is consumption of the first good, and yi is the expenditure on the second
good. Now consider the case when the unit price of the first good changes at some

threshold level of consumption, x . The household budget constraint now consists

of two segments:

p1 xi + yi if xi x

mi =
p1 x + p2 (xi x ) + yi if xi > x

where p1 is the price in the first block of consumption and p2 is the price in
the second block. The second block constraint can be simplified by substituting
mi = mi + (p2 p1 )x , so that it becomes mi = p2 xi + yi for xi > x . Graphically,
mi occurs at the intersection of the extended second segment with the y axis.

In the literature this is known as the virtual or imputed income. As price and
income are constant over each segment of the constraint each must also have
constant indirect utility value, v(p1 , mi ), and v(p2 , mi ).
The demand function is derived by initially finding the segment with the high-

est indirect utility, for which Marshallian demand is feasible. The household choice
is:

segment 1: if v(p1 , mi ) > max (v(p2 , mi ), u(x , y )) and x(p1 , mi ) < x

segment 2: if v(p2 , m) > max (v(p1 , mi ), u(x , y )) and x(p2 , mi ) > x
kink: if u(x , y ) > max (v(p1 , mi ), v(p2, mi ))

where y is the level of consumption of the numeraire that corresponds with

x , and x(p, m) is Marshallian demand. For example, segment 1 is chosen if it
has an indirect utility that is greater than that of both the other segment and the

kink, and the point of utility maximisation lies on segment 1.

Note that in the case that the budget set is nonconvex, it is possible to con-
struct a situation where there are two solutions - the utility maximising points lie
on the same indifference curve but within different segments. This can be seen in

53
 yi

budget constraint

U(x)

xi

Figure 4.1. Nonconvex Budget Constraint

Figure 4.1. In this case some additional criterion must be applied to identify a
unique choice. A solution for the nonconvex case is given by Moffitt (1986). As
we are only concerned with increasing block tariffs we confine the analysis to a
convex budget set.

With a convex budget set, the existence of a utility maximising point some-
where along a budget segment is sufficient to guarantee that segment will be
chosen. For example, consider a situation where there is a utility maximising
point strictly located on the first segment, ie. x(p1 , mi ) < x . As preferences

are strongly monotonic (alternately as utility is quasiconcave), and the budget

set is convex; both the kink and all points on the second segment must be less

54
 upper

up

p
1

upper contour set

p
2

x*

Figure 4.2. Convex Budget Constraint

preferable than x(p1 , mi ). In other words, there is an upper contour set that in-
cludes but that excludes the kink and all points on the other segment. By the
same argument, x(p2 , mi ) > x implies that no other point on the first segment
or the kink will be preferred. If there is no optimal point on either segment, then

consumption must occur on the kink. None of this precludes the possibility of a
corner solution when one of the segments is chosen.
This is shown in figure 4.2 for the two segment case.
The demand function for household i can therefore be expressed in conditional

terms as:

55

x(p1 , mi ) if x(p1 , mi ) < x





xi = x(p2 , mi ) if x(p2 , mi ) > x



x

otherwise

An equivalent compact form of demand function using indicators is given by

Moffitt (1986) as:

xi = d1 (x(p1 , mi )) + d2 (x(p2 , mi )) + (1 d1 d2 )x (4.4)

where:

d1 = 1 if x(p1 , mi ) < x otherwise, d1 = 0

d2 = 1 if x(p2 , mi ) > x otherwise, d2 = 0

4.3.2 Estimation of the IBRT Demand Function

4.3.2.1 Least Squares

The application of OLS to this model can result in biased results due to cor-
relation between the explanatory variables and the error term. In this case, the
model in 4.4 is specified with a measurement error term:

xi = d1 (x(p1 , mi )) + d2 (x(p2 , mi )) + (1 d1 d2 )x + i

As the error term increases in magnitude, the likelihood that we will observe
demand in a block adjacent to the block in which true demand occurs will also
increase. Therefore there will be positive correlation between the explanatory

variables and the error term. As noted by Moffitt (1986), the problem with the
use of OLS to estimate the demand function is caused by the incorrect assumption
that all unexplained variance in consumption is measurement error, including the

56
possibility of omitted variables. Although these assumptions hold for a large class
of problems, in the case of piecewise linear budget constraints, a more sophisticated

form of error specification is required. This lead to development of a two error

model by Burtless and Hausman (1978) for study of the effects of taxation on
labour supply, and upon which many subsequent studies have been based.

4.3.2.2 The Two Error Specification

The two error specification aims to explain all of the unexplained variation in

terms of both measurement error and heterogeneity of preferences error (HPE).

These errors have different effects. Measurement error is spread out evenly over
the budget constraint because the random component is uncorrelated with the ex-
planatory variables and across households. Heterogeneity of preferences generates

clusters around the expected points of utility maximisation. For example, given
perfect measurement of two households with identical characteristics (incomes,
size, etc.), the model would predict that utility maximisation would occur at the
same point and that consumption would be the same. The fact that it does not is
due to the inability of the model to fully account for different preferences or tastes

across households. Although this appears to be a kind of omitted variables prob-

lem, combining both measurement and heterogeneity of preferences errors leads
to biased results as just discussed. Therefore we would expect that a specifica-
tion containing two error terms with different and independent distributions will

produce substantially better estimation results.

We will consider the model with only heterogeneity of preferences error initially,
before incorporating the measurement error term. A generic form of demand
function with HPE included is:

xi = x(pi , mi , zi , , i)

57
 where xi is household i s monthly consumption of water, pi is the marginal price
paid by household i, mi is the households budget, zi is a vector of covariates, is

a vector of coefficients of the regressors (including a constant), and i is the HPE

error, assumed to have distribution: N(0, 2 ).
We will confine ourselves to a specification of HPE as an added disturbance
so the complete demand specification can be expressed as:

xi = d1 (x(p1 , mi ) + i ) + d2 (x(p2 , mi ) + i ) + (1 d1 d2 )x (4.5)

where:

d1 = 1 if x(p1 , mi ) + i < x otherwise, d1 = 0

d2 = 1 if x(p2 , mi ) + i > x otherwise, d2 = 0

The household choice conditions in 4.5 allow us to identify the dichotomous

parameters of the assumed distribution of . Segment 1 is chosen for i <

x x(p1 , mi ), segment 2 is chosen for i > x x(p2 , mi ), and the kink for
all points between. Therefore under the assumption of normality and zero mean,
these inequalities identify two critical points (, ), that define the boundaries of
choice. The amount of probability assigned to the segments and kink determines

the degree of clustering that will be evident. The distribution of i is depicted in

Figure 4.3.
The empirical evidence that observations of demand subject to piecewise linear
budget constraints tend to cluster around the kinks is supported by the theoretical

model with the assumption of normality. We can see that the support for the
kink lies between i and i and these two extremes are on either side of the
mean of the distribution. Despite the empirical evidence of clustering around
the kinks, and the two error models support for this, we note that it is possible

58
 Figure 4.3. Distribution of

to construct a piecewise linear budget constraint in which the expected point of

utility maximisation does not occur at a kink. The model can also handle this
situation. The critical points in the distribution of would be set so that the
segment probability was large and the kink probability was small.

4.3.2.3 Measurement Error

We can now reincorporate a measurement error term into the model. The

approach is to assume that measurement is normally distributed: N(0, 2 ),

and the error term is simply added to the expected demand. Furthermore we
assume that both errors are uncorrelated, ie. cov(, ) = 0, and, in general
2 6= 2 .

The model now becomes:

59
 xi = d1 (x(p1 , mi ) + i ) + d2 (x(p2 , mi ) + i ) + (1 d1 d2 )x + i (4.6)

where:
d1 = 1 if x(p1 , mi ) + i < x otherwise, d1 = 0
d2 = 1 if x(p2 , mi ) + i > x otherwise, d2 = 0

4.4 Maximum Likelihood Estimation of the Two Error Model

Maximum Likelihood Estimation (MLE) involves maximisation of a likelihood

function that is a function of the known sample values, the unknown coefficients
of the demand function, and the unknown parameters of the distributions of the
error terms. Maximisation is performed by an iterative computer algorithm that
varies the unknown coefficients and parameters in such a way as to converge to
the maxima of the likelihood function. Most econometric or statistical software

packages provide some capability in this area. With a highly nonlinear demand
function, programming the maximum likelihood procedure is generally required.
The development of the likelihood function starts with determination of the
likelihood for a single unit. Assuming that all units are independent, then the

sample likelihood is the product of the unit likelihoods, or the sum of the unit
log-likelihoods. Log-likelihood conversion is more accurate because probabilities
can be summed reducing the risk of numeric underflow.
We will initially develop a likelihood function for the two error model with a

two segment, one kink tariff. This will be followed by two proposals for adaptation
to a tariff with more than two blocks. The likelihood of a single unit is determined
by reference to a bivariate density function of the errors, for a particular value
of expected demand. The bivariate density has zero mean and contours of equal

60
probability that are elliptical in shape as determined by the variances of the errors.
For the two error demand model, maximum likelihood estimation requires finding

the set of demand coefficients that determine the errors of the data set, and the
parameters of a bivariate normal distribution that most closely fits the distribution
of these errors.

4.4.1 The Likelihood Function

The MLE approach is to reform the specification given by 4.6 in terms of the

stochastic errors whose distributions we wish to estimate. For the ith household,
the observed demand in segment 1 is given by:

xi = x(p1 , mi ) + i + i

The probability of the occurrence of this observation in segment 1 is:

Pr [(i + i = xi x(p1 , mi )) (i < x x(p1 , mi ))]

where is logical and.

If we define the total error as vi = i + i , then the contribution to the
likelihood function of an observation in the first segment is an integral over the

bivariate distribution of vi and i , that is:

Z x x(p1 ,mi )
Pr[segment 1] = h(vi , i )di (4.7)

where h is the bivariate normal distribution. Similarly, the joint probability

for segment 2 is:

Pr [(i + i = xi x(p2 , mi )) (i > x x(p2 , mi ))]

61
 and so the contribution to the likelihood function of an observation in the
second segment is:

Z
Pr[segment 2] = h(vi , i )di (4.8)
x x(p2 ,mi )

For the kink, observed demand is only subject to measurement error as the

expected demand is x :

xi = x + i

Maximisation on the kink occurs over an interval [i , i ] of centred around

its mean value. Therefore the joint probability for a kink observation is:

Pr [(i = xi x ) (x x(p1 , mi ) < i < x x(p2 , mi ))]

This may be expressed as a joint probability integrated over [i ,i ],

Z x x(p2 ,mi )
Pr[kink] = f (i)f (i )di (4.9)
x x(p1 ,mi )

The likelihood of a single observation Pr[xi ] is therefore the sum of expressions

4.7, 4.8, and 4.9:

Z x x(p1 ,mi ) Z Z x x(p2 ,mi )

Pr[xi ] = h(vi , i )di+ h(vi , i)di + f (i )f (i )di
x x(p2 ,mi ) x x(p1 ,mi )
(4.10)
This expression may be transformed so that it contains only standard normal
density and probability functions:

62
 1 1 1
Pr[xi ] = f (z1i )F (r1i ) + f (z2i )[1 F (r2i )] + f (si )[F (t2i ) F (t1i )] (4.11)
v v

where z, r, s, t are standardised forms of the errors and integration limits that

appear in 4.10 above. The complete derivation and explanation of terms is pre-
sented in the Appendix B.
Finally, the sample likelihood and the function that we wish to maximise is:

Y
L(, , ) = Pr[xi ]
i

4.4.2 Extending the Model with more than Two Blocks

In this section we formulate two different methods for extending the model
to accommodate more than two blocks. The first follows directly from the two
error model presented in Section 4.3.2.2. The second method alters the consumer

choice decision in the face of multiple blocks to a binary decision that is carried
out multiple times. Therefore preference sets will be different in both methods as
will the computed sample likelihoods.

4.4.2.1 Multiple Choice Method

We begin by extending the conditions under which the indicator variables are
set in the original model 4.6:

63
 d1 = 1 if (x(p1 , mi,1 ) + i < x1 ) otherwise, d1 = 0
d2 = 1 if (x(p2 , mi,2 ) + i > x1 ) (x(p2 , mi,2 ) + i < x2 ) otherwise, d2 = 0
d3 = 1 if (x(p3 , mi,3 ) + i > x2 ) (x(p3 , mi,3 ) + i < x3 ) otherwise, d3 = 0

where xk is the k th kink and mi,k is the virtual income in block k for household
i. These expressions can be simplified further; for example, the logical expression
in the second block becomes:

x1 < x(p2 , mi,2 ) + i < x2

x1 x(p2 , mi,2 ) < i < x2 x(p2 , mi,2 )

The kinks occupy a range of density between the segments. For example, the
second kink is between the second and third segment. Therefore the integration

limits are given by:

x2 x(p2 , mi,2 ) < i < x2 x(p3 , mi,3 )

We can now express the likelihood function for a single observation as the sum
of all s segments and s 1 kinks:

Z x1 x(p1 ,mi ) s1 Z
X xj x(pj ,mi,j )
Pr[xi ] = h(vi , i )di + h(vi , i )di +
j=2 xj1 x(pj ,mi,j )
Z s1 Z
X xj x(pj+1 ,mi,j+1 )
h(vi , i )di + f (i)f (i )di
xs1 x(ps ,mi,s ) j=1 xj x(pj ,mi,j )

64
 The modifications to 4.11 follow directly from this.

4.4.2.2 Binary Choice Method

We generalise to the n-block tariff case by considering an arbitrary kink xj .

Define LEF Tj to mean all segments and kinks to the left of xj and RIGHTj to
mean all segments and kinks to the right of xj . Following from the two segment
case in 4.7 and 4.8, the likelihood of each is:

Z xj x(pj ,mi,j )
Pr[LEF Tj ] = h(vi , i )di (4.12)

and:

Z
Pr[RIGHTj ] = h(vi , i )di (4.13)
xj x(pj ,mi,j )

Therefore the probability of the j th segment is:

Pr[segment j] = Pr[RIGHTj1] Pr[LEF Tj ] (4.14)

The kink probability is:

Pr[kink j] = 1 Pr[RIGHTj ] Pr[LEF Tj ] (4.15)

The likelihood function for a single observation is now expressed as the sum
of the first and last segments, s 2 segment product terms, and s 1 kinks:

65
 Z x1 x(p1 ,mi ) Z
Pr[xi ] = h(vi , i )di + h(vi , i )di +
xs1 x(ps ,mi,s )
s1
" Z ! !#
X Z xj x(pj ,mi )
h(vi , i )di h(vi , i )di +
j=2 xj1 x(pj1 ,mi,j1 )

s1 Z x x(pj+1 ,mi,j+1 )
X j
f (i)f (i )di
j=1 xj x(pj ,mi,j )

4.5 Chapter Discussion

In this chapter we have presented theoretical and econometric models suitable

for applied analysis of urban water consumption. The consumer problem is one
where a budget must be allocated among different goods in an optimal way. Linear

expenditure systems allows us to determine the elasticity of demand using budget

shares and price indices. The two error model is used to model the non-linear
demand that occurs when consumers are faced with an increasing block rate tariff,
an increasingly common form of urban water pricing.

66
 CHAPTER 5

Welfare and Optimal Pricing

5.1 Introduction

This chapter examines the role of welfare in water price determination. We

discuss the measurement and maximisation of welfare using conventional cost

benefit approaches. We examine the relationship between marginal cost, average
cost and economies of scale in production. Then we assess marginal cost and
second best water pricing in terms of the level of subsidy required. Second best or
Ramsey pricing has been criticised for adverse equity effects. We examine a model

of welfare maximisation that allows regulator to weight different classes of users to

achieve a more equitable outcome under Ramsey pricing. Finally we examine some
issues related to estimation of welfare using the cost and demand models presented
in the preceding two chapters. Note that in the following treatment we assume

that demand curves are formed by the horizontal summation (quantities summed
according to prices) of the household demand curves, and that the marginal cost
curve is that of a representative single utility that supplies all households.

5.2 Measurement of Welfare

The use of total net benefit as a measure of static welfare has origins dating
from the eighteenth century, but became widespread when an executive order of
the US President in 1981 mandated that regulatory arms of government adopt

67
this approach. In this section we will present a brief outline of the main result
of maximisation of total net benefit - more detail is available in texts such as

Tietenberg (1992) and Boardman, Greenberg, Vining, and Weimer (2006).

An allocation x of some good is statically efficient if its total net benefit is
maximised at a single point in time:

x = arg max (CS (x) + PS (x))

x

where consumer surplus is the excess of benefit over cost:

Z x

CS (x ) = MB (x, m)dx p x
0

and producer surplus is the excess of revenues over cost:

Z x

PS (x ) = p x MC (x)dx
0

In this notation the inverse demand or marginal benefit function is MB (x, m),
while the inverse supply or marginal cost function is MC (x), where x and m are

quantity and income respectively.

Letting (x) = CS (x) + PS (x), the maximum occurs where /x = 0, that
is:

   
MB (x) MB (x)
MB (x) MB (x) + x + MB (x) + x MC (x) = 0
x x

therefore:
MB (x ) = MC (x )

The partial equilibrium (x , p ), where marginal benefit is equal to marginal

68
 p

st
n al co
rgi
ma

consumer surplus

p*

producer surplus

ma
rgi
nal
b en
efi
t
x* x

Figure 5.1. Maximisation of Total Net Benefit

cost, is the point where total net benefit is maximised. This is shown in Figure
5.1.
In practice this equilibrium changes over time. If the good is a renewable scarce
resource such as water then a series of statically efficient allocations may deplete
the resource faster than it can be renewed. This form of scarcity can be modelled

as a constrained maximisation problem. A stream of allocations xt , t = 1 . . . T is

dynamically efficient if the sum of the net present values of the net benefit of each
allocation is maximised. We can determine the optimal allocations by maximising
the objective function:

69
 T 
X Z xt Z xt 
1t
(1 + r) MB (x)dx MC (x)dx
t=1 0 0

subject to the constraint:

T
X
qt < Q
t=0

5.3 Measurement of Welfare Change

Often it is the change (or in the simplest case the sign of the change) in

welfare that is the principal object of interest. In this section we consider the loss
of welfare arising from pricing water below marginal cost using deadweight loss as
our measure.
Deadweight loss is the loss of net benefit caused by the inefficient price. This
is:
Z x0
0
DWL(x ) = (MC (x) MB (x, m)) dx
x

where x0 is the inefficient level of output, and x0 > x . This situation is

depicted in Figure 5.2 where the deadweight loss is area C and the marginal cost

is increasing.
The contrasting cases of constant marginal cost and decreasing marginal cost
are shown in Figures 5.3 and 5.4. The dead-weight loss in both cases is area C.
These correspond respectively to constant returns to scale and increasing returns

to scale.
In Renzetti (1999) two normalised measures of inefficiency are presented. The
first is the proportional deviation from the optimal quantity of water supplied:

x0 x
DEV (x0 ) =
x0

70
 Figure 5.2. Deadweight Loss from Inefficient Pricing

The second is deadweight loss per unit of output:

waste(x0 ) = DWL(x0 )/x0

An alternative measurement of welfare change when firms supply with constant

marginal cost is compensating variation(CV) (Jehle and Reny, 2001, p. 167). This
is the change in income that is required to keep a consumers utility constant
when the price changes. Graphically, it is the area to the left of the Hicksian or

compensated demand curve bounded by the original and the new price. The CV

71
Figure 5.3. Deadweight Loss under Constant Marginal Cost

Figure 5.4. Deadweight Loss under Decreasing Marginal Cost

72
is formulated as:

Z p1
CV = xh (p, m)dp
p0

Generally Hicksian demand is not observable. In empirical work, under condi-

tions of small price changes, and when the good constitutes only a small proportion

of total expenditure, compensating variation can be approximated by the change

in consumer surplus, CS .

5.4 Welfare and Economies of Scale

We have seen that in the supply of urban water, and other regulated industries,
firms generally operate under increasing returns to scale. Therefore Figure 5.4 is
the more applicable to the case of water supply. In this case, setting the volumetric
price at marginal cost means that price is below average cost and therefore either

the price is set at marginal cost and the utility receives a subsidy to cover the
shortfall, or a Ramsey (second best) pricing scheme is adopted so the firm can
recover its costs. When the utility serves different classes of consumer, use of a
Ramsey pricing scheme creates a potential problem of equity. This is because the

markup will be higher for the class of consumers with less elastic demand, for
example low income households, compared to those with more elastic demand.
In the following three sections we discuss the relationship between economies of
scale and marginal and average cost, determine the amount of subsidy required for

strict marginal cost pricing, and lastly address the problem of equity in allocations.

5.4.1 Marginal Cost, Average Cost, and Economies of Scale

Consider a generalised cost function with m outputs and k factor prices:

73
 C = C(x1 , x2 , . . . , xm , w1 , w2 , . . . , wk )

The cost elasticity of the ith output is:

C/xi
Cxi =
C/xi

For a multiproduct firm, the measure of scale economies is defined as the

reciprocal of the sum of the cost elasticities (Kim, 1995):

1
Q= P (5.1)
Cxi

where Q is the returns to scale parameter. If outputs are priced at their

marginal cost, then:

C C
Q= P = (5.2)
pi xi R

Therefore the overall measure of scale economies is the ratio of cost to revenue.
If this is greater than one, then there are economies of scale and the utility does
not recover all its costs if marginal cost pricing is adopted.
We can also examine the relationship of economies of scale with marginal and
average cost. In the case where there is only one output we have:

AC = QMC

Therefore:

Q > 1 increasing returns to scale, AC > MC

Q = 1 constant returns to scale, AC = MC = constant

Q < 1 decreasing returns to scale, AC < MC

74
 In the case of multiple outputs we need to recognise that the portion of variable
cost devoted to production of each output is not the same as C/xi and therefore

returning to equation 5.2,

1
Q = P
(MC i xi )
AC
= P
MC i xxi


X  xi 
AC = Q MC i
x

Therefore average cost is the economies of scale multiplied by the weighted sum

of the marginal cost of each output where the weights are the share of outputs

5.4.2 Cost Recovery under Marginal Cost Pricing

Consider the amount of subsidy required if all outputs are priced at marginal
cost. This will be C R. From equation 5.2 the subsidy will be:

 
1
S =C 1
Q

For increasing returns to scale, the amount of subsidy as a proportion of cost

increases with the degree of scale economies Q. No subsidy is required when

returns to scale are constant or decreasing.

5.4.3 Equity in Ramsey Pricing

We consider now the use of a Ramsey (above marginal cost) pricing scheme
for a regulated water utility. In the case of a single output, the tradeoff for
the regulator is between decreasing consumer surplus and increasing profitability

(decreased subsidisation) of the utility. When there are multiple outputs and

75
distinct user classes for each output, problems of equity emerge as Ramsey prices
will be higher for those user classes with inelastic demand. In this section we

illustrate this with a model originally presented by Ross (1984) where the regulated
utility produces two goods that have distinct user classes, demand and prices, and
zero cross price elasticity. For a water utility, if the goods are residential and
non-residential water, then the production mix and input prices are the same and

a single marginal cost function is applicable.

The welfare problem is to maximise total consumer welfare so that the loss does
not exceed some predetermined value. In the following notation, the consumer
surplus function for good i is z(pi ), the profit function is (p1 , p2 ) and the regulator

implicitly weights the welfare of group i by i . The problem can be expressed as:
Maximise: = 1 z(p1 ) + 2 z(p2 ) subject to: > .
The Lagrangian is:

L = 1 z(p1 ) + 2 z(p2 ) ( )

The first order condition is:

z(pi ) (p)
i =
pi pi

By the definition of consumer surplus, the instantaneous rate of change in

surplus as price increases is current consumption:

z(pi )
= xi
pi

The rate of change in profit can be determined by differentiation of the revenue

and cost components R, C as follows:

76
 = R (x(p)) C (x(p))
R dx C dx
=
pi x dpi x dpi
 
dpi dx
= pi + xi MC
dx dpi

Substituting these results into the first order condition we have:

 
dpi dx
pi + xi MC = i xi
dp dpi
 
pi MC dx/dpi +
=
pi xi /pi

Therefore, expressing the own price elasticity of demand as an absolute value:

 
pi MC +
i =
pi

The left hand side of this last expression is the Ramsey number, the term

(pi MC )/pi is the markup and ranges from 0 (MC pricing) to 1. The quantity
is the shadow price of net consumer surplus - a unit increase in profit results
in a loss of net surplus of . As the Ramsey price is increased, the shadow price
of net consumer surplus decreases. As demand becomes more elastic, holding the
Ramsey price constant, the shadow price of surplus also decreases. Furthermore,

as Ramsey prices increase, the weights must be increased if elasticity and shadow
prices are held constant. These relationships suggest that the effect of Ramsey
pricing on consumer surplus and utility profit is greatest for small markups over
marginal cost but diminishes as the markup increases.

In a two good production model the ratio of weights is:

77
 1 1 M1 1
=
2 1 M2 2

Using data from other studies Resende (1997) showed that this ratio was close
to unity in the case of a sample of North American water utilities using residential
and non-residential water use to differentiate outputs. This implies that there is

no cross subsidisation of water between these groups despite non-residential users

having more elastic demand.

5.5 Estimation Issues related to Welfare Models

The figures presented in this chapter have represented marginal cost and
marginal benefit curves as linear functions of output. This implies variable economies
of scale and price elasticity over the range of output. For empirical work the use
of log-linear functional forms means that these parameters will be constant over

the range of output. This simplifies analysis considerably.

There are a variety of data deficient situations that we might encounter in
applied work when welfare measurement is the primary objective. In this section
we propose two solutions that may help overcome these problems. The first starts

with a single output unrestricted returns to scale cost function similar to the one
presented in Chapter 3. Restating this in log-linear form:

1
log C = log A + log x
Q

where Q is the returns to scale parameter. Differentiating this expression with

respect to the output price:

C p
=
p C Q

78
This suggests a functional form:

log C = log A + log p
Q

that might be useful when cost and price are observed but output is unobservable
and one of or Q is unknown.

The second situation might occur when the parameters of the consumer de-
mand function cannot be estimated because consumption data was not recorded.
This can occur, for example when households who previously used water supplied
by informal means1 , become connected to a water supply. In this case it is unlikely

that estimates of consumption under the previous system are available. However,
if prices and an estimate of elasticity are available, then an estimate of the increase
in welfare from connection to the water supply can be derived. We start with the
specification of household demand:

ln x = ln A + ln p

where A is an aggregate of covariates that is independent of price and con-

sumption and all variables are household measurements, subscripts are omitted for
clarity. Inverting and integrating this we have the Marshallian demand function:

x(p) = Ap

By definition, the increase in consumer surplus when prices are decreased from
p0 to p1 is:
1
These may include water trucks, carters, tanks, and wells.

79
 Z p0
CS = Ap dp
p1
A  1+
p0 p1+

= 1
1+

Here the original water price p0 is assumed constant for all households, while
the new utility price varies across households. Substituting for the covariates from
the demand function under the new price we have:

xp  1+
1
p0 p1+

CS = 1
1+
"  #
1+
xp1 p0
= 1
1+ p1
"  #
1+
E p0
= 1
1+ p1

where E is household expenditure on water. This approach appears in Basani,

Isham, and Reilly (2005) in the context of measurement of welfare change for
households receiving a new water connection, and in Pushpangadan and Murugan
(1998) in a study of Ramsey pricing when data limitations prevented estimation
of the demand function.

80
 CHAPTER 6

First Case Study

Urban Water Services in Victoria

Part A - Cost

6.1 The Supply of Urban Water in Victoria

Urban water in the State of Victoria is supplied by thirteen regional water

businesses and three metropolitan businesses. Each water business is operated as
a state owned company constituted by ministerial order under the Water Act 1989.
The single shareholder is the Victorian State Government. Businesses operate un-
der a board of directors appointed by the State Government with overall reporting

responsibility to the Minister for Water. The board of directors is responsible

for setting and overseeing the policies, objectives, and strategies of the business.
While they are not listed entities, water businesses have adopted the Australian
Stock Exchange Principles of Good Corporate Governance and Best Practice Rec-

ommendations, and generally follow management and reporting guidelines that

are similar to those of a listed company. Water businesses are required to pay
a dividend to the State Government based on a prescribed percentage of their
annual profit.

There are two main contractual arrangements with government. The first is a
Water Service Agreement with the Minister for Water that sets out the business
customer service obligations and quarterly reporting obligations to government.
The second is a regulatory framework termed the Water Industry Regulatory

81
Order (WIRO) established in January 2004, that sets out new regulatory ar-
rangements for water businesses. Under this framework the Essential Services

Commission (ESC) assumes an economic regulation role for the Victorian water
sector. Prices and service standards are now regulated, while each business was
required to establish a Water Plan detailing the services to be provided and pro-
posed prices for delivery of those services for the three year period commencing

July 2005.
This new regulatory regime has been introduced as a means of ensuring the
long-term sustainability of water resources. The objectives of the WIRO include
promoting the economic valuation of water rather than valuation purely on cost

of service delivery; ensuring that prices increase to reflect scarcity and user cost
of service; providing incentives for long-term investment; preventing the misuse
of monopoly power; promoting competitive market conduct; and ensuring that
consumers benefit from the gains of competition and efficiency.
The Water Plans therefore form the basis for ESCs approval of the businesss

proposed prices. Following submission of Water Plans, the ESC undertook a de-
tailed review of pricing and conducted a number of public forums seeking feedback
on the Water Plans prior to making its determinations. On 15 June 2005 the ESC
released its final decision on pricing for the years 2005/06 to 2007/08. As part of

the price review, most water businesses have now introduced a block rate tariff.
For example, the three suppliers for Melbourne are now operating under a rising
three block tariff structure.
In accordance with the need to include user costs in water services, each busi-

ness has, since 2004, collected an environmental contribution from consumers.

This is paid into a consolidated fund used for the purpose of funding initiatives
that seek to promote the sustainable management of water or address water re-
lated initiatives.

82
6.2 The Cost Data Set for Victorian Water Businesses

Each water business produces an annual report that is presented to govern-

ment, suppliers, and other interested parties. The reports contain data on con-
sumption and financial information that is the basis for preparation of a cost
data set to be used in estimation. The data set contains annual reported values

for the variables of interest for a period of four years from 2002 to 2005. The
data items extracted from annual reports are listed in 6.1. The data set is an
unbalanced panel that contains cost related data for each of the seventeen wa-
ter business (metropolitan and regional). Mergers during the period means that

some businesses have less that four years of reporting data. Dummy variables have
been introduced to account for the distinction between regional and metropolitan
business, and for time effects.
In the following sections we describe the steps undertaken to transform ac-
counting data into economic data suitable for use in the cost model.

6.2.1 Output

The output measure adopted is total volume delivered to customers. This is

a common measurement indicator reported in annual aggregate by each business.

Consumption includes both residential and non-residential (ie. business) users

but does not include major industrial consumers of water who purchase water
under wholesale contracts. Similarly agricultural users of water are excluded from
this figure. In regional areas water businesses manage their own water harvesting
and storage assets. Regional businesses operate under bulk water entitlements

that govern amounts of water that may be extracted from catchments and river
systems. In the Melbourne metropolitan area the businesses purchase bulk water
from Melbourne Water who are responsible for harvesting and storage in local

83
 TABLE 6.1

Cost and Revenue Items Identified in Financial Statements

Group Cost Item

Labour salaries, wages and employee benefits

Capital interest

Capital Adjustment capital expansion

depreciation

inventory write down

Service and Production water treatment chemicals

power, light, and water

repairs and maintenance

consultants
BOOT payments

taxes and licences

Revenue service charges

volumetric charges

government contributions

developer contributions
interest

asset sales and write downs

bulk water sales

84
catchments. Our analysis does not extend to the production and supply of this
bulk water.

Non-revenue water (NRW) is comprised of water that is lost in mains bursts,

used for fire-fighting, due to meter inaccuracies, stolen, or used internally. In
general the water businesses do not monitor NRW amounts. We do not include
other measures of output such as water quality, wastewater treated, or supply

disruptions.

6.2.2 Revenue

Revenue is not required for the cost model. However it is included in the data
set as a means of computing the average price of water. Revenue includes service
charges for water and sewerage for both residential and nonresidential customers,

volumetric charges for water (ie. those in excess of connection fees), trade waste
charges and licences and fees. The measure of revenue excludes government and
developer contributions, interest, proceeds from the sale of assets and bulk water
sales to industry. Developer contributions originate indirectly from home buyers
in new housing areas.

6.2.3 Operational Costs and Factor Prices

The dependent variable is operational costs. This includes salaries, wages

and employee benefits; interest payments; depreciation and asset write downs;
water treatment chemicals; power, light, and water; repairs and maintenance;
consultants; payments for BOOT1 services; taxes and licences. Income taxation

is not included. Overhead costs are included - these will be absorbed into the
intercept term in the cost model. For the Melbourne water businesses the cost of
bulk water is included in operational costs. Costs and factor prices are deflated
1
Build Own Operate Transfer

85
to 2002 dollars.

6.2.3.1 Capital Costs

Water business borrowings are a mix of floating, fixed interest, and non-interest

bearing debt. Businesses are required to report on their interest rate exposure
which includes determination of the weighted average effective interest rate or
weighted average cost of capital. We use this figure as our time-varying measure
of cost of capital. The determination of cost of capital is a major difficulty for cost

analysis of water utilities and there is ongoing debate regarding the use of Capital
Asset Pricing Models (CAPM) and Weighted Average Cost of Capital approaches
(WACC).
Depreciation and asset write downs is another area in which there appears to

be no standard approach in cost modelling. There is some support for including

depreciation as an operational cost, however, as pointed out in Sabbioni (2005)
this is not correct as depreciation costs are generated by investment. Furthermore,
depreciation is charged on the basis of accounting rules that may not accurately
reflect the productive capacity of capital.

We have taken the view that depreciation is a proxy for the minimum expen-
diture required to maintain the capital stock at its current level, that is, capital
expenditure allocated to renewal. Therefore assuming a constant capital output
ratio, the inclusion of depreciation means that cost observations are a lower bound

of both operating and investment costs (ie. long run cost). This compares with
the usual approach of estimating a short run operating cost function. To compare
results, we have estimated the cost function with and without depreciation. An-
other approach is the inclusion of an investment model that accounts for capital

stock deterioration and population growth; and that would be estimated simulta-
neously with supply and demand models. This would require a longer time series

86
of investment and capital expenditure data than currently available.

6.2.3.2 Labour

Annual expenditure on salaries and employee benefits and number of full time

equivalent (FTE) staff is used to derive the labour cost variable. This is deflated
to 2002 dollars.

6.2.3.3 Service Costs and Technology

We have considered a third cost factor to account for service costs such as
power, fuel, chemicals used in water treatment, and outsourced services. Out-

sourcing is increasingly common. This includes repairs and maintenance, consul-

tants, and BOOT payments. Examination of the data shows that service costs are
highly correlated with Property, Plant and Equipment (PPE) asset values. PPE
values are the written down value of assets including works in progress with ad-

justment for depreciation and asset write-downs. PPE can be treated as a proxy
for capital stock.
Technological improvements can be captured by use of a time trend variable
as a proxy for technology (Dougherty, 1992, p. 159).

6.3 Data Description

Table 6.2 and 6.3 show the variables and summary statistics for the cost data
set split into regional and metropolitan businesses. The data set consists of annual

observations for seventeen water businesses over a period of 2002-2005.

87
 TABLE 6.2

Victorian Regional Water Businesses: Summary Statistics 2002-2005

Variable Description Mean Median Min. Max. SD.

C operating costs (w/out deprecn.) 20,835 19,344 3,807 53,892 13,228

D depreciation and write downs 11,114 11,141 1,718 30,940 7,158
WL wage costs 6,532 4,893 1,100 21,756 5,201
RK borrowing costs 570 153 0 4,378 1,063
SP service costs 13,694 12,587 2,524 34,437 9,055
L labour 122.2 101.0 21.0 362.0 85.9
88

W wage rate 53 55 18 70 12
K long term debt 8,430 2,153 0 62,902 15,823
R cost of capital (WACC) 0.070 0.067 0.052 0.118 0.014
PPE property plant and equipment 347,193 324,072 65,579 892,489 231,053
X volume delivered (ML) 15,589.4 14,244.0 1,995.0 41,291.0 10,990.7
CNN connections 43,118.1 41,141.0 7,805.0 122,637.0 29,123.8
RVN revenue 25,261 24,185 4,326 72,719 16,648
p average price per KL 1.93 1.79 0.77 4.26 0.84
XPC per connection consumption 362.5 332.0 129.9 772.2 154.1

annual figures; all costs and prices are in AUD 1,000s

 TABLE 6.3

Victorian Metropolitan Water Businesses: Summary Statistics 2002-2005

Variable Description Mean Median Min. Max. SD.

C operating costs (w/out deprecn.) 215,419 222,538 142,108 266,728 38,037

D depreciation and write downs 25,567 28,532 12,427 32,707 6,975
WL wage costs 22,669 21,994 14,039 30,270 6,071
RK borrowing costs 24,002 21,967 11,011 35,644 8,832
SP service costs 35,391 35,539 20,295 47,675 8,113
L labour 358.5 374.0 218.0 430.0 76.2
89

W wage rate 63 63 51 76 7
K long term debt 393,154 364,607 191,161 576,769 135,910
R cost of capital (WACC) 0.061 0.060 0.058 0.063 0.002
PPE property plant and equipment 1,101,813 1,163,784 669,045 1,372,657 240,600
X volume delivered (ML) 144,401.1 147,419.5 108,800.0 170,143.0 21,753.2
CNN connections 515,790.1 563,601.1 296,055.0 594,000.0 113,841.4
RVN revenue 268,571 283,224 179,229 311,230 46,830
p average price per KL 1.86 1.90 1.64 2.13 0.18
XPC per connection consumption (KL) 288.0 287.5 232.0 369.9 45.6

annual figures; all costs and prices are in AUD 1,000s

6.4 Estimation of Two Factor Production and Cost Models

In this section we present several variants of production and cost models that
have been described in Chapter 3. The models are estimated as log-linear models
with correction for heteroskedasticity based on a weighted least squares procedure
(Pindyck and Rubinfeld, 1991, p. 129).

Dummy variables are included to account for differences between Metropolitan

(Melbourne) and regional businesses, and time effects. All models have been
estimated with cost and price data items expressed in $000s, and all flows are
annual.

6.4.1 Two Factor CRS Production Function

The two factor CRS production function with dummy and time variables is:

ln(Xit /Lit ) = ln(A) + ln(Kit /Lit ) + 1 CLASSi + 2 Tt

where:

X = volume delivered (ML)

K = plant, property and equipment
L = number of full time equivalent staff
CLASS = dummy variable 0:metropolitan water business 1:regional water business

T = time variable: number of years since 2002

i = firm index
t = time index

Table 6.4 shows the estimated coefficients for the basic production model and

model variants with CLASS and time dummy variables.

90
 Standard errors are shown below the estimates in brackets. Variables that are
significant at the 95% level are marked with an asterisk beside the standard error.

The righthand columns show the adjusted R2 value and the standard error of the
residuals.
The adjusted R2 parameter indicates a increasing level of fit for as dummy
variables are added to the model. All variables are significant. The sign of the

CLASS dummy variable indicates that relative to regional water businesses, the
metropolitan water businesses exhibit economies of scale. The coefficient on the
time dummy variable also indicates some decrease in output over time. The
most probable explanation of this is reduced output/consumption due to drought,

water restrictions and increased prices; while inputs remained at the same level.
Although this indicates some inefficiency, both capital and labour are more sticky
than consumption and this result is consistent with expectations.
The output elasticity for K is the coefficient for K/L (ie. ) and the output
elasticity for L is 1 . The general result indicates a considerably higher output

elasticity for capital than for labour; this suggests that future increases in output
will occur through increased investment in productive capital.

6.4.2 Two Factor Cost Function with CRS Production

The two factor CRS minimum cost function with CRS production dummy and
time variables is:

   
Cit rit
ln = ln(A) + ln + 1 CLASSi + 2 Tt
wit Xit wit

where:

91
 TABLE 6.4

CRS Production Function

Model const ln(K/L) CLASS T R2

basic -2.16 0.899

(1.25) (0.155)* 0.361 1.91

class -0.10 0.761 -1.1969

(1.21) (0.151)* (0.0844)* 0.824 2.15

class, time -0.778 0.8651 -1.1864 -0.0855

(0.694) (0.0858)* (0.0698)* (0.0350)* 0.860 1.99

C = operating cost
X = volume delivered
w = wage price
r = weighted average cost of capital

CLASS = dummy variable 0:metropolitan water business 1:regional water business

T = time variable: number of years since 2002
i = firm index
t = time index

Table 6.5 shows the estimated coefficients for the constant returns to scale cost
model exclusive and inclusive of depreciation. The model variants are based on
the inclusion of CLASS and time dummy variables. The coefficient for r/w (ie. )
is the output elasticity for K in the Cobb-Douglas production dual. The output
elasticity for L is 1 . In the restricted model the output elasticities are also the

cost shares. The coefficients for the dummy and time variables are additive to the
scaling constant ln(A).

92
 TABLE 6.5

CRS Cost Function

Model const ln(r/w) CLASS T R2

depreciation excluded

basic -1.046 0.3827

(0.514) (0.0806)* 0.271 2.35

class -0.368 0.4798 -0.0657

(0.662) (0.0960)* (0.0810) 0.372 2.38

time -0.877 0.4412 0.1353

(0.337) (0.0564)* (0.0289)* 0.516 2.14

class, time -0.879 0.4401 0.0359 0.1129

(0.517) (0.0759)* (0.0739) (0.0282)* 0.502 2.25

depreciation included

basic 0.182 0.512

(0.832) (0.128)* 0.206 2.18

class -0.292 0.4782 0.2805

(0.675) (0.0975)* (0.0822)* 0.622 2.10

time 0.193 0.5531 0.1513

(0.266) (0.0457)* (0.0199)* 0.743 2.27

class, time -0.131 0.5316 0.2465 0.1292

(0.471) (0.0687)* (0.0722)* (0.0286)* 0.730 2.47

93
6.4.3 Two Factor Cost Function with Variable Returns to Scale

The two factor minimum cost function with variable returns to scale produc-
tion, including dummy and time variables is:

   
Cit 1 wit
ln = 0 + ln (Xit ) + ln + 1 CLASSi + 2 Tt
rit + + rit

where:

C = operating cost

X = volume delivered
w = wage price
r = weighted average cost of capital
CLASS = dummy variable 0:metropolitan water business 1:regional water business

T = time variable: number of years since 2002

i = firm index
t = time index

Table 6.6 shows the estimated coefficients for the variable returns to scale cost

model exclusive and inclusive of depreciation. The model variants are based on the
use of CLASS and time dummy variables. In the model the estimated coefficients
for ln(w/r) are the cost shares for L; and the inverse of the coefficient for ln(X)
determines returns to scale. The output elasticities can be determined from the
coefficients. The coefficients for the dummy and time variables are additive to the

scaling coefficient ln(A).

94
 TABLE 6.6

Variable RTS Cost Function:

Model const ln(X) ln(w/r) CLASS T R2

depreciation excluded

basic -0.484 0.9233 0.645

(0.728) (0.0319)* (0.107)* 0.942 2.61

class 2.12 0.7965 0.5009 -0.478

(1.13) (0.0580)* (0.0853)* (0.177)* 0.972 2.20

time -0.505 0.9406 0.590 0.1262

(0.672) (0.0275)* (0.102)* (0.0356)* 0.960 2.20

class, time 1.84 0.7954 0.521 -0.461 0.0766

(1.15) (0.0566)* (0.111)* (0.170)* (0.0284)* 0.971 2.16

depreciation included

basic 1.584 0.8344 0.5212

(0.579) (0.0311)* (0.0765)* 0.931 2.27

class 2.62 0.7857 0.4632 -0.203

(1.11) (0.0563)* (0.0839)* (0.177) 0.960 2.01

time 1.435 0.8901 0.414 0.1505

(0.712) (0.0336)* (0.104)* (0.0413)* 0.931 3.22

class, time 3.21 0.8109 0.315 -0.246 0.0827

(1.21) (0.0606)* (0.107)* (0.175) (0.0280)* 0.971 2.23

95
6.5 Discussion of Results

Overall the variable returns model performed better than the constant returns
model. Based on the adjusted R2 parameter, a superior level of model fit was
achieved. The inclusion of dummy variables also improved the fit of the constant
returns model. The most likely explanation of this is that the coefficient of the

log of output is constrained to one in the constant returns models meaning that
more of the variance in cost is explained by other factors, raising their significance.
Based on F statistics shown in Table 6.8, every model variant was significant, ie.
the null hypothesis of all slope parameters equal to zero was rejected in every case.

For both models, all of the explanatory variables are significant, with the
exception of the CLASS dummy variable which is only significant in the constant
returns model when depreciation is included and in the variable returns model
when this is not the case. The inclusion of depreciation in cost does not appear
to improve the significance of the explanatory variables in any consistent manner.

The sign of the explanatory variables (output and factor prices) are in all cases
positive and conform to expectations. The sign of the CLASS dummy variable in
each model points to different evidence in respect of costs between regional water
and the metropolitan water businesses. For fixed input prices and output, the

models do not reveal a consistent pattern of higher or lower costs for one class of
business over the other. All cost models show a consistent increase in costs over
time, holding other factors constant. All cost variables were adjusted for inflation,
therefore this result indicates an increase in real costs over time, a not unexpected

result.
The cost elasticities are the coefficients of the log linear function
log C = f (log X, log r, log w). These are equal to the cost shares. Using duality,
the output elasticities of the cost minimising production function are determined

96
from the factor price parameter estimates of the cost function. The returns to
scale parameter is the inverse of the coefficient of output in the variable returns

model. These estimated values appear in Table 6.7.

Output elasticities (changes in factor inputs) for capital are generally higher in
the variable returns models with depreciation included, matching the production
model results. The constant returns models do not show any consistent difference

in output elasticities based on the inclusion or otherwise of depreciation. Cost

elasticity for changes in output indicate the presence of economies of scale (in-
creasing returns to scale) in all variants of the variable returns model. A test of
the hypothesis for constant returns to scale as opposed to increasing returns to

scale was rejected at the 95% level. This is detailed in the next section. The cost
elasticity for labour prices is higher than for capital when depreciation is excluded
but lower when depreciation is included. This indicates the sensitivity of results
to the inclusion of depreciation, and suggests that long run costs (in the variable
returns model) are more sensitive to interest rates than to wage rates. This pro-

vides some explanation as to why, under previous pricing arrangements, there has
been limited capital investment by water businesses.
Finally, as shown in Table 6.9, we have carried out Regression Specification
Error Test (RESET) tests for model misspecification using artificial models aug-

mented with squared and cubed predicted values (Hill, Griffiths, and Judge, 2001,
p. 187). In all but one case we were unable to reject the null hypothesis that
the coefficients of the augmented regressors were both zero. This suggests that
functional form and omitted variables are not a problem in these models and is

supportive of correct model specification.

97
6.5.1 Testing for Constant Returns to Scale

Based on the regression results of the variable returns to scale models we

would like to determine if the returns to scale parameters shown in Table 6.7 are
significant. That is we would like to confirm that the Victorian water businesses

exhibit increasing returns to scale. This has been carried out with a series of
hypothesis tests, the results of which are summarised in Table 6.10. Test statistics
are derived from the estimates and standard errors in the variable RTS model,
equation 6.6.

In all cases the null hypothesis of constant returns to scale was rejected at the
95% level. In the case where the alternative was increasing returns to scale, the
null hypothesis was also rejected. Therefore there is statistical evidence that the
Victorian water businesses exhibit increasing returns to scale.

98
 TABLE 6.7

Summary of Estimation Results: Victorian Cost Data

Output Elasticity Cost Elasticity

Model capital labour output int.rate wage rate

CRS Cost - excluding depreciation

basic 0.383 0.617 1.0 0.383 0.617

class 0.480 0.520 1.0 0.480 0.520
time 0.441 0.559 1.0 0.441 0.559
class, time 0.440 0.560 1.0 0.440 0.560

CRS Cost - including depreciation

basic 0.383 0.617 1.0 0.383 0.617

class 0.480 0.520 1.0 0.480 0.520
time 0.553 0.447 1.0 0.553 0.447
class, time 0.440 0.560 1.0 0.440 0.560

Variable RTS - excluding depreciation

basic 0.384 0.699 0.923 0.355 0.645

class 0.626 0.629 0.797 0.499 0.501
time 0.436 0.627 0.941 0.410 0.590
class, time 0.602 0.655 0.795 0.479 0.521

Variable RTS - including depreciation

basic 0.574 0.625 0.834 0.479 0.521

class 0.683 0.589 0.786 0.537 0.463
time 0.658 0.465 0.890 0.586 0.414
class, time 0.845 0.388 0.811 0.685 0.315

99
 TABLE 6.8

F Tests for Model Significance

H0 : 2 = 0, 3 = 0, . . . , k = 0 Model is of no significance
H1 : at least one of the i 6= 0

Model Test Statistic DF F95% Decision

CRS Cost - excluding depreciation

basic 22.5 1,57 4.0 reject H0

class 18.2 2,56 3.2 reject H0
time 31.9 2,56 3.2 reject H0
class, time 20.5 3,55 2.8 reject H0

CRS Cost - including depreciation

basic 16.1 1,57 4.0 reject H0

class 48.8 2,56 3.2 reject H0
time 84.7 2,56 3.2 reject H0
class, time 53.3 3,55 2.8 reject H0

Variable RTS - excluding depreciation

basic 473.5 2,56 3.2 reject H0

class 684.2 3,55 2.8 reject H0
time 459.9 3,55 2.8 reject H0
class, time 483.1 4,54 2.5 reject H0

Variable RTS - including depreciation

basic 390.1 2,56 3.2 reject H0
class 467.5 3,55 2.8 reject H0
time 260.9 3,55 2.8 reject H0
class, time 493.7 4,54 2.5 reject H0

100
 TABLE 6.9

Ramsey RESET Tests for Model Specification Error

original model yi = 0 + 1 x1,i + . . . + k xk,i

predictions yi = b0 + b1 x1,i + . . . + bk xk,i
artificial model yi = 0 + 1 x1,i + . . . + k xk,i + 1 yi2 + 2 yi3

H0 : 1 = 0, 2 = 0 (no specification error)

H1 : at least one of the i 6= 0

Model DF Test Statistic (Ft ) Pr(F > Ft ) Decision

CRS Cost - excluding depreciation

basic 2,55 0.0963 0.908 do not reject H0

class 2,54 0.4092 0.666 do not reject H0
time 2,54 0.3851 0.682 do not reject H0
class, time 2,53 0.1986 0.821 do not reject H0

CRS Cost - including depreciation

basic 2,55 0.1759 0.839 do not reject H0

class 2,54 0.8538 0.431 do not reject H0
time 2,54 0.3047 0.739 do not reject H0
class, time 2,53 0.5368 0.588 do not reject H0

Variable RTS - excluding depreciation

basic 2,54 4.4291 0.017 reject H0

class 2,53 0.1020 0.903 do not reject H0
time 2,53 2.2409 0.116 do not reject H0
class, time 2,52 0.0068 0.993 do not reject H0

Variable RTS - including depreciation

basic 2,54 1.0526 0.356 do not reject H0

class 2,53 0.0351 0.966 do not reject H0
time 2,53 0.3102 0.735 do not reject H0
class, time 2,52 0.0342 0.966 do not reject H0

101
 TABLE 6.10

Hypothesis Tests for Constant Returns to Scale

Alternative is either increasing or decreasing RTS

H0 : 1/( + ) = 1 H1 : 1/( + ) 6= 1
DF: 57 critical value: 2.003

Model Variant Test Statistic Decision

basic -2.406 reject H0

class -3.510 reject H0
class, time -3.612 reject H0

Alternative is increasing RTS

H0 : 1/( + ) = 1 H1 : 1/( + ) < 1
DF: 57 critical value: 1.673

Model Variant Test Statistic Decision

basic -2.406 reject H0

class -3.510 reject H0
class, time -3.612 reject H0

102
6.6 Applying a Translog Cost Function to Analyse the Impact of Service Costs

In this section we present the results of estimation of a system of cost share

equations derived from a constant returns to scale translog cost function. The
derivation of the equations has been described in Chapter 3 and follows the
methodology presented in Greene (2003, p. 368) and originally applied in a study

of US manufacturing industries by Berndt and Wood (1975).

We have augmented the data set with an additional factor: services, so that
a system of three share equations can be estimated. Service costs include energy,
use of contractors and consultants, and production costs such as bulk water and

treatment chemicals. The objective in this case is estimation of short run oper-
ational costs, therefore depreciation has been excluded as a cost. Two methods
are used to determine a price for service costs. The first is a value price that
is the service cost per unit of capital employed. The second approach uses the
Service Industries Producer Price Index published quarterly by the Australian

Bureau of Statistics (2007). The use of price indices in lieu of market price data
is appropriate in contexts where an aggregate of goods contributes to the relevant
factor. In general, price indices are used when a system of share equations is to
be estimated, see for example Dachraoui and Harchaoui (2004).

The system of share equations is:

shw = w + ww log w + wr log r + ws log s

shr = r + wr log w + rr log r + rs log s

shs = s + ws log w + rs log r + ss log s

Where w, r, s are factor prices of labour, capital and service respectively. The
third equation is eliminated and all prices expressed as ratios of the service price

103
to avoid singularities. The system has been estimated using the SystemFit soft-
ware package (Henningsen and Hamann, 2006). There are six parameters to be

estimated and one restriction (the third parameter of the first equation and the
second parameter of the second equation). Results are presented in Tables 6.11
and 6.12.
The signs of the estimates of the coefficients of the share equations are con-

sistent regardless of the service price used. They indicate several points. Firstly
an increase in the wage rate means that the cost share of labour increases while
the cost share of services falls. An increase in the cost of capital produces a fall
in borrowing share and an increase service cost share. An increase in service

price produces a fall in labour cost share, increased borrowing cost share, and in-
creased service cost shares. The results suggest that labour is sticky and not easily
substituted by services, while in contrast services are more easily substituted for
capital.
The own price elasticities have the expected signs and show a higher price

elasticity for capital than for the other factors. This is supported by the low debt
to equity ratios of most water businesses, particularly regional ones (Victorian
Water Industry Association, 2005). Notable also is the greater demand elasticity
of services in comparison to labour per unit cost of capital increase, and the high

demand elasticity of services in response to a unit labour cost increase. Both of

these suggest that there is a tendency to increase the use of services in response
to increases in capital and labour costs.

104
 TABLE 6.11

Share Equation Estimates: Translog Cost Function

Services Value Price Services Price Index

cost share log w log r log s log w log r log s

labour -1.008 0.184 0.006 -0.190 0.521 0.091 0.021 -0.112

borrowing 0.022 0.006 -0.055 0.050 -0.345 0.021 -0.054 0.033
services 1.986 -0.190 0.050 0.140 0.824 -0.112 0.033 0.079

TABLE 6.12

Estimates of Elasticities: Translog Cost Function

Services Value Price Services Price Index

w r s w r s

w -0.077 w -0.395
r 1.472 -2.323 r 2.765 -2.291
s 0.026 2.836 -0.122 s 0.424 2.222 -0.214

105
 CHAPTER 7

First Case Study

Urban Water Services in Victoria
Part B - Marginal Cost and Welfare

7.1 Introduction

Part A of this case study presented the results of estimation of linear Cobb
Douglas (constant and variable returns to scale) and translog models of the cost
function for the Victorian Water Businesses. Variants of these models employed a
time dummy variable and a CLASS dummy variable that distinguished between

metropolitan and regional water services. We observed an improvement in model

fit from the constant returns to the variable returns models. The hypothesis
of constant returns to scale was rejected in favour of the alternative that water
production exhibits increasing returns to scale. The cost elasticity of output was

estimated to be in the range 0.786 to 0.941.

This part of the case study contains two main contributions. Firstly, we present
several methods for determining the marginal cost of urban water based on the
estimated cost function. These are classified by the selection of output level at

which cost is determined and by the underlying model. Secondly, we examine the
welfare effects associated with current price levels and our marginal cost estimates.

106
7.2 Estimating Marginal Cost

7.2.1 Constant Returns and Variable Returns Models

We define the conditional marginal cost function as the marginal cost function

using mean values of the factor prices. The conditional marginal cost is the con-
ditional marginal cost function evaluated at some representative value of output.
The constant returns model yields a conditional marginal cost function that is a
constant while the variable returns model yields a function that is nonlinear in

output.
The following shows how the conditional marginal cost is determined for the
variable returns model with a CLASS dummy variable. Because we include the
CLASS dummy variable the cost functions for metropolitan and regional author-
ities will differ in the scale parameter, as will the factor price sample means r and

w.
We first restate the cost function from 3.4 as a conditional cost function:

1
C = exp(A)w Q r Q x Q

where, for clarity A = 0 + 1 CLASS from the model specification, and Q is

the scale economies parameter. Differentiation with respect to output yields the
conditional marginal cost function for which we take logs and simplify:

1 1
MC = exp(A)w Q r Q x Q 1
Q    
  1 1
log(MC ) = A + log w Q r Q + log + 1 log(x)
Q Q




1


Hence substituting B = A + log w Q r Q + log Q
, we have the conditional

107
marginal cost function:

 
1
log(MC ) = B + 1 log(x) (7.1)
Q

Marginal cost estimates are determined by evaluation of this function at factor

price means for some specified level of output.

7.2.2 Partial Equilibrium Marginal Cost

In the previous section we used means of factor prices and output to determine
the marginal cost. To contrast this we now determine marginal cost using a partial
equilibrium of the marginal cost equation and a log-linear demand equation. The
advantage of this approach is that the equilibrium (welfare maximising) level of

output is also selected, avoiding the need for choosing the level of output at which
to measure marginal cost.
We adopt a simple model of system demand with price as the only explanatory
variable:

log x = b0 + log p

Substituting this into equation 7.1 and equating marginal cost and price, we

have the marginal cost at the point of equilibrium between demand and supply:

1


B+ Q
1 b0
log(MC ) = 1

(7.2)
1 Q
1

7.2.3 Translog Function Marginal Cost

In Part A of this Case Study we estimated the parameters of the translog

average cost function for the Victorian Water Businesses. This was estimated
as a three factor system without dummy variables. The estimation results were

108
presented in Table 6.11. In this section we evaluate this function at output and
factor means to form a conditional translog average cost estimate.

In the original share equation system, the estimate of the intercept term 0
is not available because the average cost equation is excluded from the system.
This estimate is required to compute marginal cost. Therefore, the system was
augmented with the original average cost equation and re-estimated. Price ratios

were again used to avoid singularities. The services value price was used as the
numeraire. The estimates produce predicted average cost for any vector of prices.
To determine aggregate average cost we have fitted the sample predicted values
to a normal distribution and taken the mean of that distribution as the average

cost.
The translog system is:

pw pr 1 pw 1 pr pw pr
log g = 0 + w log + r log + ww log2 + rr log2 + wr
ps ps 2 ps 2 ps ps ps
pw pr
shw = w + ww log + wr log
ps ps
pw pr
shr = r + wr log + rr log
ps ps

The estimated coefficients of the system differ slightly to the original share
equation system because of the extra equation. For the depreciation excluded
sample the equations are:

pw pr pw pr pw pr
log g = 2.647 0.950 log 0.036 log + 0.088 log2 0.030 log2 + 0.014
ps ps ps ps ps ps
pw pr
shw = 0.950 + 0.176 log + 0.014 log
ps ps
pw pr
shr = 0.036 + 0.014 log 0.059 log
ps ps

109
 For the depreciation included sample the equations are:

pw pr pw pr pw pr
log g = 1.947 0.591 log + 0.345 log + 0.056 log2 + 0.053 log2 0.008
ps ps ps ps ps ps
pw pr
shw = 0.591 + 0.112 log 0.008 log
ps ps
pw pr
shr = 0.345 0.008 log + 0.105 log
ps ps

7.3 Discussion of Results

The three methods presented in the preceding section have been used to es-
timate marginal cost at a particular level of output. Table 7.1 summarises these
results. The columns of this table are in order:

CLASS: The model and data apply to all businesses, only metropolitan, or

only regional businesses.

Fn. Evaluation: The method used to evaluate the marginal cost function.

Output (KL): The value of annual water delivered used in evaluation.

no depn./with depn.: These columns show the marginal cost based on data

sets exclusive and inclusive of depreciation.

VP: The average volumetric charge for this class of water businesses.

Table 7.2 shows the sample means used in calculations. All prices and costs

have been deflated to 2002 prices.

As expected, the different models yield different results. The constant returns
translog model yielded higher marginal costs relative to the other models. The

110
 TABLE 7.1

Marginal Cost Results: Victorian Water Authorities

MC. ($/KL)

CLASS Fn. Evaluation(*) Output (KL) No Depn. Depn. VP.(**)

CES - Constant Returns to Scale

combined sample means (FP) 37,422 1.49 2.14 0.72

metropol. sample means (FP) 206,710 1.55 1.70 0.80
regional sample means (FP) 21,765 1.43 2.20 0.70

Translog - Constant Returns to Scale

combined sample means (FP) 37,422 1.96 0.74 0.72

combined mean of predicted cost 37,422 1.75 2.52 0.72

CES - Increasing Returns to Scale

combined sample means (FP+OP) 37,422 1.29 1.58 0.72

metropol. sample means (FP+OP) 206,710 1.17 1.28 0.80
regional sample means (FP+OP) 21,765 1.12 1.68 0.70

combined partial equilibrium 17,290 1.34 1.80 0.72

metropol. partial equilibrium 198,148 1.07 1.19 0.80
regional partial equilibrium 12,827 1.08 1.74 0.70

* FP=factor price; OP=output

** source: Victorian Water Industry Association (2005)

111
 TABLE 7.2

Sample Means of Parameters: Victorian Water Authorities

Factor Variable Sample Metro Region

cost of capital r 0.067 0.061 0.070

wage rates ($000s) w 54.721 62.940 53.021

output (ML) x 37,421.9 206,709.5 21,764.5

connections H 123,232 515,790 43,118

per connection output (ML) x 0.350 0.288 0.363

price ($/KL) p 0.72 0.80 0.70

variable returns model produced consistently lower estimates than the other mod-

els. The translog model would be expected to model the variation in factor prices
better than the others as it allows second order effects compared to linear or ex-
ponential models that are constant or consistently decreasing. However, because
the model assumes constant returns to scale, the slope of the cost function is con-

strained to be constant and does not capture the effect of increasing returns to
scale (marginal costs decreasing) for higher levels of output.
The differences in model results are more striking when depreciation is taken
into account. With depreciation included, the variable returns model shows a

higher marginal cost for regional businesses over metropolitan businesses. An

examination of the data shows that for regional businesses the cost share of de-
preciation was 33% while for metropolitan businesses it was 10%. For the period
for which we have data there has been considerable depreciation and writeoff of
assets by regional authorities, relative to their metropolitan counterparts. The

percentage of capital expenditure directed towards renewal of assets in 2004/2005

112
was 40% for regional businesses, compared to 25% for metropolitan businesses
(Victorian Water Industry Association, 2005). All models have captured this dif-

ference quite clearly. This suggests that the inclusion of depreciation in addition
to capital expenditure is necessary to capture long run costs related to investment.
As mentioned in Chapter 3, there is little uniformity in other applied studies in
respect of the treatment of depreciation. These estimates indicate the inclusion

of depreciation as a cost shifts the marginal cost curve upwards - without sig-
nificantly changing its shape, or cost elasticity of output as was shown in Table
6.7.
The results when a system demand function is used to determine the equilib-

rium value of output do not substantially differ from the increasing returns model
that sets output at the sample mean. The efficient level of output for regional
water businesses is considerably lower than the sample data. This suggests that
current consumption levels could be lowered in regional areas with a corresponding
reduction in marginal cost.

We make two final observations regarding the marginal cost estimates. First,
the disaggregated estimates (into metropolitan and regional water businesses) are
preferable to the combined estimates because the sample output means are ap-
propriate to that class of business. This suggests that size is an important factor

in this type of analysis and needs to be incorporated at each step, from function
estimation to pricing. Second, we can observe that all estimates are substantially
higher than the prevailing average volumetric charges. This supports the generally
accepted belief that water is under-priced.

7.3.1 Data Aggregation and Sensitivity Analysis

This work highlights the use of cost and demand functions in marginal cost
determination - in contrast to conventional industry practice that is reliant on

113
present value estimates based on historical records. As a result of the need to
transform a functional form into a scalar quantity, the marginal cost estimates

produced by the above described procedure depend on the mean values of factor
prices and output. This is a common approach in applied work, the mean values
used form part of the assumptions of the results. An alternative approach would be
to leave the choice of representative values to the decision maker by presenting the

results in the form of a two-dimensional plot of marginal cost against individual

factor prices or output. If the former approach is adopted and functions are
evaluated at mean values of their parameters we expect that the results will be
sensitive to those mean values. A second source of dependency relates to the

values of the coefficient estimates. These are reported in the regression analysis
as scalars but of course a parameter estimate is only the most statistically likely
one out of an entire distribution.
In this section we carry out sensitivity analysis on the sample mean values
and parameter estimates that form the basis of marginal cost estimates. As we

have estimated a number of models under differing conditions, it is not realistic to

conduct sensitivity analysis on every model variant. Therefore we have restricted
this analysis to an examination of the sensitivity of the variable returns model to
changes in its parameters and mean value of output. This choice was also driven

by the observation that marginal cost is not very sensitive to the value of output
- indicating low economies of scale in production.
We have calculated lower and upper values of the marginal cost with coeffi-
cient estimates and mean output changed by one half of their standard error -

a standard form of sensitivity analysis. We need to emphasise that these values

are not confidence intervals, rather the range of each indicates the sensitivity of
the model to changes in the parameter. Table 7.3 shows the results of estimation
of lower and upper values of conditional marginal cost for the variable returns

114
 TABLE 7.3

Sensitivity Analysis of Marginal Cost Estimates: Variable Returns Model

Lower/Upper MC (parameter 0.5 std. err.)

parameter combined metropol. regional

w/out depreciation
intercept 0.90 1.86 0.66 2.06 0.64 1.98
1/Q 1.08 1.56 0.80 1.71 0.82 1.54
0.91 1.85 0.87 1.57 0.85 1.49
output X 1.24 1.41 1.15 1.19 1.06 1.23

with depreciation

intercept 1.18 2.11 0.73 2.23 0.96 2.92

1/Q 1.32 1.90 0.88 1.85 1.23 2.28
1.22 2.04 0.96 1.71 1.27 2.21
output X 1.45 1.91 1.26 1.30 1.57 1.84

model using mean output. Sensitivity analysis was carried out on the intercept,

the output elasticity estimate , the cost elasticity estimate 1/Q (the coefficient
of output), and output level. A combined and disaggregated estimate is shown
and the analysis has been conducted on both the depreciation excluded data set
and the depreciation included data set.

The main observation we can make in respect of this sensitivity analysis is

that the largest range in marginal cost estimates occur in the intercept (ie. of the
log-linear MC function). This is not unexpected as the intercept shifts the entire

marginal cost curve. Furthermore, because the intercept and the coefficients are
exponents, the value of marginal cost will be very responsive to these estimates.
This is a feature of all log-linear models that is seldom considered in applied work.

115
The choice of output level is the least sensitive as a result of the scale economies
being close to one. The lower range of estimates approximate existing prices.

These results are confirmatory of the sensitivity of parameter estimates. It is

interesting to note that use of a conventional polynomial cost function means that
the marginal cost will be invariant to the intercept, in contrast to an exponential
(log-linear) form where the intercept still remains after differentiation.

7.4 Measurement of Deadweight Loss and Required Subsidy

As indicated in Table 7.1, average prices are less than the marginal cost es-

timates that we have obtained from a variety of cost models. Starting in 2005,
annual price increases are being awarded to Victorian water authorities on the
basis of Water Plans submitted to the ESC. The motivation for price increases is
to reduce consumption and cover investment costs in the near term.

In Chapter 5 we presented the methodology for calculating the deadweight

loss arising from pricing below marginal cost based on Renzetti (1999), and the
measure of additional subsidy required for increasing returns to scale production
when marginal cost pricing is employed.

deadweight loss: loss of welfare at current prices

R x0
DWL(x0 ) = x
(MC (x) MB (x)) dx

deviation: deviation from efficient output

DEV (x0 ) = (x0 x ) /x0

waste: average DWL per unit of current output

waste(x0 ) = DWL(x0 )/x0

subsidy required at efficient price with increasing returns to scale

116
 1


S =C 1 Q

The initial output level is x0 is determined from the demand function at the

current price p0 . The efficient point (x , p ) lies at the intersection of the demand
and marginal cost curves. It is the solution to the system:

log x = 0 + log p

log p = 0 + 1 log x

where is the price elasticity of demand.

The results are shown in Table 7.4. Two additional metrics are shown: the
subsidy required per unit of output (at the efficient point), and the subsidy re-

quired per unit of output as a percentage of the efficient price. The former is a
measure of the amount the price would need to increase by to achieve full cost
recovery.
We comment firstly on the results using the depreciation excluded data set.
This shows that the regional water businesses are less efficient in terms of the per-

centage deviation from efficient levels of output and the waste (the average DWL
per unit of output). A higher proportional increase in price would be required
for regional consumers than for those in metropolitan areas. At the efficient level
of output, both metropolitan and regional water businesses would require similar

levels of subsidisation of around 30% of the efficient price to fully recover costs.
With the depreciation included data set the results are more striking. Re-
call that the inclusion of depreciation was motivated by the need to account for
investment costs over a period that has been characterised by low levels of invest-

ment in Victorian water supply. Depreciation was considered to be a proxy for

117
 TABLE 7.4

Welfare Measurements: Variable Returns Model

parameter metropolitan regional

w/out depreciation

price 2004/05 ($/KL) 0.80 0.70

qty. demanded 2004/05 (ML) 272,140 26,243
efficient price ($/KL) 1.08 1.08
efficient output (ML) 215,024 18,703
dead weight loss (K$) 7,164 1,320
deviation (pct) 21.0 28.7
waste ($/ML) 26.32 50.30
subsidy at efficient price (K$) 59,129 5,168

subsidy per unit output ($/KL) 0.27 0.28

subsidy (pct of price) 29.6 29.9

with depreciation

price 2004/05 ($/KL) 0.80 0.70

qty. demanded 2004/05 (ML) 272,140 26,243
efficient price ($/KL) 1.19 1.75
efficient output (ML) 198,171 12,829
dead weight loss (K$) 13,334 6,959
deviation (pct) 27.2 51.1
waste ($/ML) 49.00 265.16

subsidy at efficient price (K$) 64,543 6,114

subsidy per unit output ($/KL) 0.33 0.48
subsidy (pct of price) 38.9 83.3

118
infrastructure renewal costs. In the period for which data was available, propor-
tionally higher levels of depreciation (and renewal investment) were undertaken by

regional water businesses in comparison to their metropolitan counterparts. The

effect of this is an increase both in total cost and in marginal cost. This explains
the higher price for regional businesses at the efficient level of output, and the
sharp drop in consumption (as a result of the unchanged demand curve), and the

higher deviation measure. The waste is also significantly higher than the without
depreciation case because the renewal investment cost has not been reflected in
the current price. Similarly, because costs are higher the required subsidy is higher
even when efficient pricing is carried out.

7.5 Conclusion

There is considerable informal and empirical evidence that the economic cost
of water is not being fully accounted for in the provision of water services. This
case study has explored a functional approach to determining the marginal cost of

water supply for a sample of Victorian Water businesses covering the period 2002-
2005. Based on this sample there is evidence that the Victorian Water Authorities
marginal cost curves exhibit increasing returns to scale, or decreasing marginal
and average costs. Evaluation of marginal cost functions at the means of factor

prices and outputs indicate that prices generally are below marginal costs. The
use of a translog function produces estimates of marginal cost that are higher than
with log-linear models. The effect of including depreciation in cost is to shift the
marginal cost curve upwards. There is a loss of welfare as a result of inefficient

pricing and this is higher in average cost terms for regional water businesses than
for metropolitan businesses. As water businesses operate under increasing returns
to scale, a subsidy is still required if marginal cost pricing at efficient levels of
output were to be undertaken. This subsidy increases when capital costs are

119
taken into account.
A functional approach to marginal cost determination creates the problem of

choice of factor price and output levels. In this case study we have explored
several alternate solutions to this. We expect that functions need to be adjusted
as variables change. Under regulatory pricing however, tariff changes generally
lag behind changes in the underlying input costs and demand. We will discuss

the possibility of more frequent price adjustments in the conclusion to this thesis.
Another issue for the use of marginal cost to determine price is related to the case
where there are increasing returns to scale. In this case, full cost recovery is not
possible and therefore a form of subsidy or second best pricing is required.

The period for which data was available comes before a period, commencing
in 2005, of considerable transformation in water management in Victoria. This
has involved increasing the role of the regulator in price determination. Prices
are expected to increase by up to 8% on the basis of authorities water plans that
set out planned capital investment projects over the next five years. Therefore as

investment increases so will the long run marginal cost.

Our present understanding of the long run marginal cost of urban water supply
is dependent on the cost reporting of utilities. This study has identified three areas
in which reporting falls short of the requirements to understand these costs.

Firstly, a significant proportion of operational and investment costs born by

the water authorities is in the form of outsourced services and BOOT projects.
The costs associated with these services tend to be amortised over time and appear
as fixed costs to the authority. As these costs are not accounted for in marginal

output terms, and not disaggregated into labour, capital and energy inputs and
prices, there is a loss of information concerning their influence on marginal costs.
Secondly, through developers, new home buyers pay connection fees and for a
part of the network infrastructure costs. The fees are accounted for as revenues

120
by the water businesses and form a considerable component of their income. This
means that purchasers of houses in new residential developments are subsidising

low water prices for everyone by paying for the network expansion. This is one
reason that house prices are high in Australia. On the other hand, if the true cost
of network expansion was borne by all in the form of higher water prices, then
owners of established dwellings would be subsidising the infrastructure required

to support new housing development at the fringes of our major cities.

Thirdly, reporting of costs by water authorities does not fully account for
marginal societal or user costs. An environmental contribution charge has recently
been introduced, but the actual assessment of the costs to which this charge applies

is informal and qualitative, and depends on the initiative of individual authorities.

121
 CHAPTER 8

Second Case Study

Urban Water Services in Manila

Part A - Cost

8.1 The Privatisation of Water Services in Manila

When the city of Manila was faced with critical water shortages during 1995,
the Government passed the Water Crisis Act (Chotrani, 1999; Rosenthal, 2001),
establishing the case for privatisation of the water supply. The public run system
was in decay. Non-revenue water (NRW), the proportion of supplied water lost to
leakages and illegal taps, was among the highest in the world. The Metropolitan

Waterworks and Sewerage System (MWSS), a division of the Department of Public

Works, was considered inefficient (Montemayor, 2003), and the cost of operation
of the citys water supply was a significant contribution to the national budget
deficit.

With support from the World Bank, the objectives of privatisation were laid
out. These included increased efficiency in operation and management of the
water supply, reduced government debt from interest payments and staff costs,
the opportunity for a return on assets while retaining government ownership of

major infrastructure, and increased investment in the water system leading to

improved services to all residential and commercial customers.
The bidding process involved companies competing on the basis of investment,
coverage, NRW reduction, and water quality. However, the six short listed com-

122
panies competed ultimately on price - that being the weighted average price of
water that they would be offering to their residential and commercial consumers.

In August of 1997 the Government handed over responsibility for management of

water and sewerage in the greater Manila region to two private companies, Manila
Water Company Inc. (MWCI) and Maynilad Water Supply Inc. (MWSI). The
city was split into an East and a West Zone, with each company responsible for

one zone. MWCI successfully bid for the East Zone with an average price of PhP
2.32 ($0.08) per cubic metre1 . This was 26% of the average price before privati-
sation. MWSI was successful in bidding for the West Zone with a price of PhP
4.96 ($0.17) per cubic metre, 56% of the rate before privatisation. Table 8.1 shows

price and volumes before and since privatisation.

Under the concession agreement, the MWSS handed over its operational roles
to the concessionaires and established a Regulatory Office (MWSS-RO). The reg-
ulators responsibility was to monitor pricing, contract compliance, and perfor-
mance, and to promote competition. The MWSS retained ownership of the stor-

age and network assets. The new concessionaires would maintain and invest in
the network, manage the billing systems, and pay monthly concession fees to the
government. Tariffs within each zone would be set independently by the conces-
sionaires, but any changes to these tariffs could only be allowed by the Regulatory

Office according to strict guidelines.

In the initial period of operation consumers benefited from lower tariffs while
the concessionaires commenced a programme of investment that was aimed at
reducing NRW and increasing revenues. With the Asian Economic Crisis in late

1997 there was, however, a dramatic increase in the debt servicing cost for both
concessionaires. This lead to insistence from both concessionaires that a mech-
1
The currency is the Philippine Peso. At the time of privatisation the exchange rate was
PhP 29.33 = USD1.

123
anism be incorporated in the tariff that would allow them to recover from cus-
tomers the foreign currency losses that occurred as a result of Peso devaluation.

In 2001 this was agreed to by the government and a tariff Currency Exchange
Rate Adjustment (CERA) mechanism was introduced. This resulted in a series of
extraordinary (ie. above CPI) price adjustments that have forced the government
to defend its privatisation strategy and respond to increasing criticism over the

cost of water and lack of adequate service, mostly from groups that represent the
urban poor, but also from industry and the middle class. As shown in Table 8.1,
since privatisation, the nominal average price of water has increased at an average
annual rate of 20%.

Considerable domestic debate has been entered into regarding the bid process
(Solon and Pamintuan, 2000), and subsequent performance of the companies in-
volved in Manilas water privatisation. According to its critics, in many areas
the service had not improved, infrastructure investment had not materialised, and
there were not the promised number of new connections. The strategy of reducing

NRW by more effective monitoring of illegal connections was not as successful as

hoped. These themes are revisited frequently by the local media and domestic
NGOs, see for example, Esguerra (2003). The price of water is a matter of great
sensitivity in Manila - a number of NGOs exist purely for lobbying congress on

water pricing issues and preventing local distribution monopolies in areas not ser-
viced by the concessionaires. More recently however, the outlook for one of the
concessionaires has become more promising. During 2005 MWCI was listed on
the Philippine stock exchange and, with a new source of investment capital, is

reporting significant decreases in NRW2 .

2
MWCI reported a fall in NRW from 59% to 37% in 2005.

124
 TABLE 8.1

Metro Manila Water Tariff and Production 1996-2003

(Tariff PhP/m3 )
Date Event Basic All-In Connections Prodn.(MLD)

MWSS
01-Aug-96 pre-privatisation 6.12 8.62 779,380 2,800.00
01-Aug-97 on privatisation 8.78 11.83 779,380 2,800.00

MWCI
01-Aug-97 award of contract 2.32 4.02 310,682 1,542.00
01-Jan-99 CPI 2.61 4.37 332,582 1,668.34
01-Jan-00 CPI 2.76 4.55 339,491 1,689.50
01-Jan-01 CPI 2.95 4.78 352,982 1,724.00
21-Apr-01 CPI 3.22 5.11 352,982 1,724.00
21-Oct-01 AEPA 4.22 6.32 352,982 1,724.00
01-Jan-02 FCDA 6.75 9.37 369,699 1,662.91
01-Jan-03 rebasing 10.06 13.38 396,778 1,577.83

MWSI
01-Aug-97 award of contract 4.96 7.21 449,234 1,864.36
01-Jan-99 CPI 5.80 8.22 498,051 2,177.13
01-Jan-00 CPI 6.13 8.62 547,880 2,250.77
01-Jan-01 CPI 6.58 9.17 577,637 2,417.42
21-Apr-01 CPI 6.58 9.17 577,637 2,417.42
21-Oct-01 AEPA 10.79 14.26 577,637 2,417.42
01-Jan-02 FCDA 11.39 19.92 573,194 2,362.85
01-Jan-03 rebasing 11.39 19.92 585,953 2,313.82

125
 TABLE 8.1

Continued

Notes:

Basic weighted average price for residential and commercial users

All-In basic price plus charges, levies and taxes
CPI annual consumer price index increase
AEPA accelerated extraordinary price adjustment
FCDA foreign currency devaluation adjustment

Sources: MWSS-RO, McIntosh and Yniquez (1997)

Republic of Philippines (2002)
Republic of Philippines (2004)

Pressured by government, both concessionaires have instigated programs for

low cost water supply to households in economically depressed parts of the city,

where the connection coverage is low or zero and where the population is de-
pendent on more expensive suppliers such as water tankers. In the East Zone
this scheme is called Tubig para sa Barangay, and in the West Zone it is called
Bayan-Tubig; both literally translate to Water for the Community. These are low

cost distribution schemes that provide service to multiple households from a single
connection. These programmes have the potential to significantly reduce illegal
taps and revenue lost to other suppliers who often source their water stocks from
MWSS at cheap commercial rates and sell them at greatly inflated prices. In some

areas of the city, public taps have been removed and the new low cost schemes pro-
vided some compensation for this. However many households still cannot afford
the one-off connection fee and for that reason do not participate in the schemes
although it would mean lower prices. Areas that are zoned for redevelopment

126
within the next five years have missed out on these schemes because the water
companies are unable to recover their investment costs within this period (Perez,

2003).
These positive accomplishments have been offset somewhat by a long-running
dispute between Maynilad Water, the concessionaire for the West Zone, and the
MWSS-RO. On March 8, 2001, Maynilad Water (MWSI) served a Force Majeure

notice on MWSS and stopped paying its monthly concession fees of PhP 200M
(US4M). Concession fees are the rental cost for the water supply and distribution
network, and also are a means of the concessionaires contributing to development
costs of new infrastructure projects that were committed to by the government

before privatisation. Maynilad cited the effects of the El Nino drought from 1997
to 1998, and the failure of MWSS to complete vital infrastructure projects on
time including the Umiray Angat Transbasin Project (UATP). When this project
was finally completed on June 23, 2000, Maynilad claimed that the supply was
insufficient to meet its requirements.

On 9 December 2002, MWSI sought to terminate its 25 year concession early

by filing a Notice of Early Termination, citing breaches of the Concession Agree-
ment by the MWSS. At the same time they sought the return of their US$120M
performance bond and PhP 19 billion (about $365M) in compensation from Gov-

ernment. The consortiums foreign partner, the French firm Suez Lyonnaise des
Eaux, withdrew. Subsequently, an arbitration panel declared on 7 Nov 2003, that
there were no grounds for early termination, the concession agreement had to
continue in force, and the parties were to find extra-judicial solutions to their

problems. The panel further declared that the overdue concession fees, which had
grown to PhP 6.77 billion by September 2003, were still payable. MWSI continued
to service the West Zone during this dispute.
Under threat of MWSS being able to draw on the performance bond, Maynilad

127
agreed to a Rehabilitation Plan that was approved in September 2004. Subsequent
non-execution of this plan lead to Maynilad creditors executing a Debt Capital

and Restructuring Agreement (DCRA) that involved write-off of advances and

dilution of the parent companys equity interest. Under the DCRA, MWSS is
given the right to subscribe to a majority (83.97%) of the shares of Maynilad
Water and to assign any portion of those shares to a third party by way of a

public offer. Benpres, the parent company of Maynilad, divested its equity in the
firm on July 20, 2005.
These events highlight some of the risks involved in privatisation and the need
for governments to ensure that private operators of essential services remain fi-

nancially viable.

8.2 Production and Costs

We now turn our attention to the study of production and costs under the
privatised system.

Metro Manilas water supply is sourced from the Angat, Ipo and La Mesa
dams, to the north of the city. Angat is the main source, water is channelled
directly into Ipo dam and then through a number of aqueducts that discharge
into the Novaliches Reservoir and onto the two treatment plants. The Balara

Treatment Plant serves the eastern part of the city (serviced by MWCI) and has a
production capacity of 1,600 million litres daily (MLD). The La Mesa Treatment
Plant serves the western half of the city (MWSI) and has a production capacity of
2,400 MLD. Combined, the plants service more than six million people throughout

the metropolis. The concessionaires each have additional storage facilities with
total combined storage capacity of 460 million litres.
The concessionaires annual aggregate production and costs are taken from
annual reports, Security and Exchange Commission (SEC) filings, and Service

128
Performance Reports (SPR) produced monthly and annually for the Regulatory
Office by each concessionaire. This data is available for the period from the

commencement of the concession in August 1997 up to December 2005. The

Philippine financial year is calendar based.

In the following paragraphs we discuss the steps undertaken to transform ac-

counting data into economic data suitable for use in the cost model.

8.2.1 Output

There are several choices of output measurement. These include water pro-
duced before transfers or losses occur; water net of transfers; and water net of
transfers and losses. Transfers between concessionaires were needed in the ini-

tial years of operation while some infrastructure works were completed but are
no longer needed and have been phased out. A transfer can be considered profit
neutral in the sense that it appears as a cost for one firm and revenue for the
other. Water loss through leakage and pilfering (NRW) is lost revenue, but is not

itemised as a cost by the concessionaires. Instead, reduction of NRW3 is a measure

of performance that is monitored by each firm and the MWSS-RO. We consider
declining NRW as a technical efficiency gain brought about by investment that
leads to increased net water. For the concessionaires then, water net of transfers

and losses (available water ) is the correct measure of output as it determines the
revenue that the firm will receive. Therefore this is the measure of output we
have used. As we will estimate a single output model, secondary measures such as
water quality, waste water treated, or supply disruptions have not been included.
3
NRW is measured as a percentage of available water.

129
 TABLE 8.2

Manila Cost Data: Cost Items Identified in Financial Statements

Category Cost Item

Labour salaries, wages and employee benefits

Capital and Investment interest

foreign exchange losses
concession fees
interest on performance bond
depreciation and amortisation
provision for inventory write down

Production power, light, and water

water treatment chemicals
repairs and maintenance
waste water costs
transfer costs

Managerial regulatory fees

management, technical and professional fees
business meetings and representation
postage, telephone and supplies
advertising
occupancy costs
provision for doubtful accounts
collection fees
taxes and licences
transportation and travel
insurance

130
8.2.2 Operational Costs and Factor Prices

The dependent variable is operational costs. This is the sum of variable input
costs and overhead costs but excludes capital expenditure and depreciation. All
costs are deflated to 1997 Pesos. The following sections first discuss overhead costs

followed by each of the variable input costs.

8.2.2.1 Overhead Costs

These include occupancy (office rental) costs, regulatory costs, and fees for
managers and consultants. Regulatory costs equal to one-half of the annual MWSS
budget are paid by each concessionaire and serve to cover the operational costs

of the MWSS-RO. These are capped (at PhP 200.00 million in 2005) and are
adjusted annually according to the CPI. Fees for managers and consultants are
incurred in project management activities. For example, an infrastructure devel-
opment project requires technical and managerial specialists for the duration of

the implementation. The cost will be related to the total project cost, that is to
the incremental level of capital stock, and this expansion is dependent on the cost
of capital.

8.2.2.2 Labour

The measure of labour input was computed from staff per connection ratios
contained in Annual and Service Reports. The price of labour was computed

from total wage costs in the annual income statements. This compares favourably
with both the June 2002 Occupational Wages Survey (Republic of Philippines,
2003) and daily nominal minimum wage rates (Republic of Philippines, 2005).
The Occupational Wages Survey shows the average monthly wage for workers

engaged in the collection, purification and distribution of water in the utilities,

131
water treatment and related industries to be PhP 20,161. Average annual staff
salaries have been deflated to 1997 Pesos.

8.2.2.3 Cost of Capital

We adopt the usual approach of considering capital to be the plant and equip-
ment assets that are used in production. For the Manila concessions, ownership of
these assets is distributed across both firms and the government. The concession-
aires pay concession fees for those assets that are government owned and allocate

additional funds for new investment4 . Concession fees are calculated as a propor-
tion of MWSS historical and current debt at the commencement of the concession
in August 1997. This proportion depends on the location of the underlying assets
- approximately 90% was located in the West Zone at the commencement of pri-

vate operation. For the concessionaires the cost of capital is therefore the cost of
their portion of the MWSS long term debt plus the cost of their own short and
long term borrowings.
Over the period 1997-2005 for which we have data, both firms have been re-
quired to borrow to finance their business. The capital structures of both firms

are a complex mix of debt and equity. Equity is sourced from parent companies
and foreign investors. The debt of both companies is a mix of fixed and variable
interest rate loans, and some small interest free loans. Loans are primarily denom-
inated in USD, and to a lesser extent in Pesos and Yen. The benchmark interest

rate for offshore borrowing is the six month London Interbank Overnight Rate
(LIBOR) and for domestic borrowing it is the 364 day Treasury Bill rate. At the
end of 2005, MWCIs long term debt amounted to US $84.7 million (PhP 4,526
million) and MWSIs long term debt was approximately US $115.7 million (PhP
4
Under the concession agreement these investments will be returned to the government at
the end of the 25 year concession.

132
6,179 million). The two firms account for their concession fees as a long term debt
obligation and include concession assets in their own balance sheet reporting.

Newly borrowed funds have been utilised for two main purposes. Initially
they have been used to fund the concession startup and the first few years of
operation until the concession became profitable. The second purpose has been to
invest in the supply network with the objective being to increase revenues through

increasing the number of connections and decreasing water losses. Increases in

water output are attributable to growth in service coverage5 through increased
connections rather than directly from population growth. Before privatisation,
service coverage was around 67%, by 2003 this had grown to 85% (MWCI) and

78% (MWSI).
There are several approaches to modelling the cost of capital. One alternative
used by Wolak (1994) employs a user cost of capital approach. This is determined
as a time series autoregressive equation using as regressors the regulators allowed
rate of return, the change in prices of capital goods, and their depreciation. We

have adopted a simpler approach based on available data that involves calculation
of a weighted cost of capital (WACC). The weighted cost of capital of each firm can
be calculated for each time period with knowledge of interest rates for each type of
loan and the loan amount. We have simplified this by determining the proportion

of total debt denominated in $US and assuming the remainder is denominated in

Peso. Therefore the WACC is:

rt = t rtU S + (1 t )rtP eso

Where the t is the proportion of $US debt and r are the respective interest
rates in period t.
5
The proportion of households with a connection.

133
8.2.2.4 Energy Costs and Technology

Energy, chemicals, and water itself are inputs to the production of water.
Electricity usage costs have been reported on by one concessionaire but not the
other; neither supplier reports the unit prices of these inputs. As there are no

suitable proxies for these factor prices they have been omitted from the analysis.
We have already mentioned that declining levels of NRW can be viewed as
a technological improvement that leads to productivity gains. Other forms of
technology improvements might include new treatment technologies that reduce

costs. There is unlikely to be a proper time series measure of such technology.

The favoured approach is therefore to use a time trend variable as a proxy for
technology.

8.2.2.5 Environmental Costs

Environmental (user or society costs) are increasingly incorporated into eco-

nomic models to account for the effects of human impact on the environment.
The Philippines government has begun to address this by use of an Environmen-
tal Charge (EC) of 10% of the water charge that is levied upon water consumers.
This is collected by the concessionaires as revenue. This charge is used to offset
waste water costs and government imposed environmental levies or licences. The

part that is returned to government is used by environmental agencies in rehabil-

itation and environmental law enforcement. Therefore there is no effect on cost
minimisation choices.

8.2.2.6 Foreign Exchange Losses and Gains

Foreign exchange losses and gains are incurred due to currency movements im-

pacting monthly amortised loan repayments in local currency. The local currency

134
differential arising from a loss or gain is added to a foreign currency adjustment ac-
count and recorded as an accounting cost. The net adjustment account is reviewed

every six months by the regulator and concessionaire to determine the adjustment
to the Foreign Currency Differential Adjustment (FCDA) that is charged to cus-
tomers as a recovery mechanism. This is in addition to the permanent Currency
Exchange Rate Adjustment (CERA) that is levied on consumers at the rate of

one Peso per cubic metre consumed. The mechanism allowing recovery of these
costs was only introduced in October 2001, four years after commencement of the
concessions. Accumulated foreign currency losses are amortised as expenses at
the same rate that the FCDA allows recovery. Because of the delay in allowing

these costs, and the need to amortise them, there is lag between the causal event
- devaluation of local currency - and its cost recording. An examination of the
cost data clearly shows the increase in costs commencing late 2001 and into 2002
as a result of this recording.
Therefore foreign exchange costs are recouped as lagged revenue. They will

not enter the firms short run production decisions but may do so in the capital
adjustment process in particular when macroeconomic factors are impacting in-
terest rates. However, knowing that consumers will pay all currency losses reduces
the incentive for the firms to hedge against such losses.

8.3 Data Description

Table 8.3 shows the variables and summary statistics for the cost data set.
Additional ratio variables used in estimation have not been included. The data

set consists of annual observations for the two concessionaires over the period of
1997-2005.

135
 TABLE 8.3

Manila Cost Data: Summary Statistics

Variable Description Mean Median Min. Max. SD.

C Annual Operating Costs 2,005.36 1,966.50 973.56 3,488.62 881.74

(M.Peso)

X Available Water (MLD) 740.245 754.500 440.000 877.431 109.019

136

W Average Annual Wage Rate 272.243 264.529 252.430 298.163 17.701

(K.Peso)

RWACC Cost of Capital (WACC) 5.45 5.74 1.56 8.39 2.17

RDOM Domestic Interest Rate 12.03 10.9 8.9 18.4 3.13

LIBOR London Interbank Rate 4.00 3.77 1.23 6.65 1.95

EXCHRATE Exchange Rate 46.84 50.99 29.47 56.04 8.75
8.4 Estimation of the Minimum Cost Models

In this section we present the estimation results of both the constant returns
to scale cost model and the variable returns to scale cost model that have been
described in Chapter 3. A translog model was not estimated because it would
require more observations to produce reliable results. The models are estimated

in log-linear form with weighted least squares correction for heteroskedasticity.

Unobserved time effects are accounted by use of an additive time variable which
alters the scaling constant. We use duality to retrieve the parameters of the
production function and have not estimated the direct production function. All

local currency values have been deflated to 1997 Pesos.

The models account for cost of capital by use of a weighted cost of capital
(WACC) approach computed as the weighted average of US dollar borrowings
and Peso borrowings. Based on the available data, it was preferable to express
this in terms of the proportion of total debt denominated in USD. The WACC

can then be expressed as:

W ACCt = t LIBORt + (1 t )RDOMEST ICt

where:

W ACCt = weighted average cost of capital in time t

t = proportion of debt denominated in USD
LIBORt = the London Interbank Overnight Rate6
RDOMEST ICt = domestic interest rate
6
The benchmark for foreign currency borrowings.

137
8.4.1 Two Factor Cost Function with CRS Production

The two factor CRS minimum cost function with CRS production dummy and
time variables is:
   
Cit rit
ln = 0 + ln + 1 Tt
wit Xit wit

where:

C = operating cost
X = available water
w = wage price
r = weighted average cost of capital

T = time variable: number of years since 1997

i = firm index
t = time index

Table 8.4 shows the estimated coefficients using pooled data for the basic cost

model and model variants with a time variable. The coefficient for r/w (ie. )
is the output elasticity for K in the Cobb-Douglas production dual. The output
elasticity for L is 1 . In the restricted model the output elasticities are also the
cost shares. The coefficients for the dummy and time variables are additive to the

scaling constant 0 . Standard errors are shown below the estimates in brackets.
Variables that are significant at the 90% level are marked with an asterisk beside
the standard error. The rightmost columns show the adjusted R2 value and the
standard error of the residuals.

138
 TABLE 8.4

Two Factor Cost Function with CRS Production

Model Variant const ln(r/w) T R2

basic 2.926 0.184

(0.621) (0.160) 0.019 1.37

time 3.039 0.204 -0.0040

(0.706) (0.202) (0.0377) 0.020 1.29

8.4.2 Two Factor Cost Function with Variable Returns to Scale

The two factor minimum cost function with variable returns to scale produc-
tion, including a time variable is:

   
Cit 1 wit
ln = 0 + ln (Xit ) + ln + 1 Tt
rit + + rit

where:

C = operating cost
X = available water

w = wage price
r = weighted average cost of capital
T = time variable: number of years since 1997
i = firm index

t = time index

Table 8.5 shows the estimated coefficients using pooled data for the basic cost
model and model variant with time variable. In the model the estimated coeffi-

cients for ln(w/r) are the cost shares for L; and the inverse of the coefficient for

139
 TABLE 8.5

Two Factor Cost Function with Variable RTS Production

Model Variant const ln(X) ln(w/r) T R2

basic 1.21 1.317 0.734

(3.48) (0.550)* (0.133)* 0.753 1.39

time -7.72 2.572 0.975 -0.0926

(6.51) (0.958)* (0.203)* (0.0661) 0.739 1.50

ln(X) determines returns to scale. The output elasticities can be determined from
the coefficients. The coefficients for the dummy and time variables are additive

to the scaling coefficient 0 .

This model is a significant improvement on the constant returns to scale model
with higher adjusted R2 values and all variables significant with the exception of
the time variable.

8.5 Discussion of Results

Of the two models that we have estimated, the variable returns to scale model
produces the better fit as indicated by the adjusted R2 value. Excluding the time

variable, all variables are significant at the 90% level. Based on F -test statistics
shown in Table 8.7, the constant return model was not significant, ie. the null
hypothesis of all slope parameters equal to zero could not be rejected. On the other
hand F -tests indicate the variable returns model was significant in all parameters.

Using the results from the variable returns model we can determine the output
elasticities and cost shares. These are summarised in Table 8.6. For comparision
we have included the CRS results as well although these are not statistically

140
significant.
We can make two comments regarding these results. The first is that there

appears to be some evidence that the output elasticity of labour is higher than
for capital. We would conclude from this that water production is relatively
labour intensive. The second observation that we make is that there is evidence
of decreasing returns to scale (in the variable returns model). Decreasing returns

to scale means that both marginal and average costs are increasing, and marginal
costs are greater than average costs. If water were to be priced at marginal
cost then the firms revenue would be in excess of its costs (Luenberger, 1995,
p. 63), leading to potential criticism by consumers and the regulator. On closer

examination however, as will be shown in the following hypothesis test, this result
cannot be confirmed statistically. We cannot reject the hypothesis that returns to
scale are constant, that is that marginal costs and average costs are constant and
that the firms make normal profits.
Finally, as shown in Table 8.8, we have carried out Regression Specification

Error Test (RESET) tests for model misspecification using artificial models aug-
mented with squared and cubed predicted values. Because the constant returns
model did not pass the F -test we only carried out RESET tests on the variable
returns model. In both variants we were unable to reject the null hypothesis that

the coefficients of the augmented regressors were both zero. This suggests that
functional form and omitted variables are not a problem in the variable returns
model.

141
 TABLE 8.6

Summary of Estimation Results: Manila Cost Data

Output Elasticity Cost Share

Model Capital Labour Capital Labour RTS

CRS production 0.184 0.816 0.184 0.816 1.0

Variable RTS production 0.202 0.558 0.266 0.734 0.760

TABLE 8.7

F Tests for Model Significance

H0 : 2 = 0, 3 = 0, . . . , k = 0 Model is not significant

H1 : at least one of the i 6= 0

Model Test Statistic DF F95% Decision

Constant Returns Model

basic 1.3 1,16 4.5 do not reject H0
time 1.2 2,15 3.7 do not reject H0

Variable Returns Model

basic 27.0 2,15 3.7 reject H0
time 17.0 3,14 3.3 reject H0

142
 TABLE 8.8

Ramsey RESET Tests for Model Specification Error

original model yi = 0 + 1 x1,i + . . . + k xk,i

predictions yi = b0 + b1 x1,i + . . . + bk xk,i
artificial model yi = 0 + 1 x1,i + . . . + k xk,i + 1 yi2 + 2 yi3

H0 : 1 = 0, 2 = 0 (no specification error)

H1 : at least one of the i 6= 0

Model DF Test Statistic (Ft ) Pr(F > Ft ) Decision

Variable Returns Cost Model

basic 2,13 0.7617 0.487 do not reject H0
time 2,12 1.9124 0.190 do not reject H0

8.6 Testing for Constant Returns to Scale

We have carried out a series of hypothesis tests to determine if the returns

to scale parameters shown in Table 8.6 are significant. The results of these tests
are summarised in table 8.9. Test statistics are derived from the estimates and

standard errors in the variable returns to scale model 8.5. In both cases we do not
reject the null hypothesis of constant returns to scale at the 90% level. Therefore
there is statistical evidence that the Manila concessionaires water production
exhibits constant returns to scale.

143
 TABLE 8.9

Hypothesis Tests for Constant Returns to Scale

Alternative is either increasing or decreasing RTS

H0 : 1/( + ) = 1 H1 : 1/( + ) 6= 1

DF: 16 critical value: 1.746

Test Statistic Decision

0.576 do not reject H0

Alternative is decreasing RTS

H0 : 1/( + ) = 1 H1 : 1/( + ) > 1

DF: 16 critical value: 1.337

Test Statistic Decision

0.576 do not reject H0

144
 CHAPTER 9

Second Case Study

Urban Water Services in Manila
Part B - Demand

9.1 Urban Water Consumption and Prices in Manila

9.1.1 Sources of Data

Over the past decade there have been several studies related to domestic wa-
ter use carried out in Manila. The most detailed survey of domestic water con-
sumption was carried out in 1996 by the Philippine Institute for Development

Studies (David, Inocencio, Abracose, Clemente, and Tabios, 1998) with the objec-
tive being determination of the economic price of water. This survey covered 500
households in the Metro Manila area within the National Capital Region1 (NCR).
Another large survey was conducted in 2000 by the MWSS Regulatory Office and

the World Bank. This was a survey of MWSS water users (PAWS-Public Assess-
ment of Water Services) that aimed to determine consumers perceptions of the
benefits of privatisation and concessionaires performance. The concessionaires,
private consulting firms, and development assistance donors such as the ADB are
all potential sources of data, but this data is not available for research purposes.

For this study we have made use of two large data sets consisting of housing
census and household expenditure information for estimation of parameters of the
1
The Philippines is divided into regions. The NCR covers Manila and some outer districts.
Manila itself consists of a number of cities, some of which are classed as municipalities.

145
demand function. This requires the use of household water expenditure and prices
to derive consumption. The advantage of these data sets, in addition to the fact

that they are in the public domain, is that they are random samples covering a
wide geographical area, thereby reducing the risk of selection bias. In the following
two sections we describe our chosen sources of information.

9.1.1.1 The 2000 Census of Population and Housing

The 2000 Census of Population and Housing (National Statistics Office, 2000b)

was undertaken by the National Statistics Office (NSO) in May 2000. It was
the eleventh census of population and the fifth census of housing undertaken in
the Philippines since the first census in 1903. The objective of the census is to
take an inventory of population and housing units all over the Philippines and

to collect information about their characteristics. Census day was May 1, 2000.
Enumeration lasted for about one month. The housing component of the census
(hereafter referred to as the Housing Census) is a randomly selected 10% sample of
29,686 households that were required to complete additional questions concerning
housing characteristics. The variables that are of interest in the Housing Census

concern the household source of water for drinking and cooking, and for laundry
and bathing.

9.1.1.2 The 2000 Family Income and Expenditure Survey

The NSO conducts the Family Income and Expenditure Survey (FIES) every
three years. The objectives of this survey are to gather data on family income

and expenditures, sources of income and income distribution, spending patterns,

and the degree of inequality among families; to update weights used for estima-
tion of the Consumer Price Index; and to assist in estimation of the national
poverty threshold and incidence. The data collected are not used for taxation,

146
investigation or enforcement purposes. The expenditure data are grouped into
food and non-food expenditure. This latter category includes expenditure on do-

mestic water, bottled water, power and lighting, durables, housing rental and
maintenance, tobacco and alcohol, clothing, medical needs, transport and com-
munications, recreation, education, and gifts. Income data are grouped by source.
The survey also includes data related to family size and composition.

A national sample of 39,615 households was surveyed for the FIES. The pri-
mary survey division is a domain. A domain is an urban or rural administration
unit (city, municipality, or district). Rural areas with population over 150,000
(based on the 1995 National Census) are also domains. The households were

interviewed twice using the same questionnaire over a six month period. Each
interview used the previous half-year period as the reference period. According
to the NSO, this scheme improves data quality by reducing errors in survey re-
sponses and averages seasonal variation in income and expenditure. The first
interview was in July 2000 while the second was in January 2001. The survey

utilises a multi-stage sampling design for sample frame selection; details of which
are provided in National Statistics Office (2000a).

9.1.2 Household Sources of Water

Households in Manila get their domestic water for drinking, cooking, bathing
and laundry from different sources including the MWSS network, wells, roof tanks,

and water carters. The choice of water supply depends on a number of factors
including location, dwelling type and household income. Many households have
alternative water sources, for example a roof tank and MWSS connection. Table
9.1 below shows the division of households by primary source of water in the

National Capital Region (NCR).

The cost of a shared connection is divided equally or according to use among

147
 TABLE 9.1

Source of Household Water in Manila NCR

% of All Households
Source of Household Water Households Drinking/Cooking Laundry/Bathing

MWSS - single connection 1,083,072 50.78 51.50

MWSS - shared connection 518,091 24.29 24.20

Well - private/shared 317,591 14.89 16.78

Carter/Tanker 135,205 6.34 5.30

Bottled water 27,603 1.29 0.00

Spring lake river rain etc 3,629 0.17 0.20

Others 47,798 2.24 2.00

Total Households 2,132,989 100.00 100.00

Source: NSO 2000 Housing Census.

the households that share it. The account holder (this might be one family or the
manager of a housing estate) will apportion the monthly account and may add a
small service charge to each household. A well may incur a charge for the user if
it is not on their land. The government is also seeking to register all wells and to

charge households for groundwater extraction.

9.1.3 Water Prices and Household Income

Based on a survey of 500 households in Metro Manila conducted in 1996,

David, Inocencio, Abracose, Clemente, and Tabios (1998) found that monthly
water expenditure consumed around 2.4% of household income and 11.9% of per

148
capita income for the most expensive service; private water carters selling direct
to the household dwelling, a form of water source prevalent amongst the lowest

income earning households. This secondary market is often characterised by local

monopoly suppliers who, as reported in the study, source about 80% of the water
from the MWSS network, and then distribute it at much higher prices throughout
the poorer parts of Manila. Households have no option other than to pay these

prices because they lack substitutes for essential domestic water Those house-
holds with the best form of water service, MWSS with a sewer, had the lowest
ratio of water expenditure as proportion of income at around 0.3%. This survey
demonstrated the inverse relationship between income and the average price of

water. The household budget share for water is significantly higher for lower in-
come groups simply because these groups live in areas that are not serviced by
the water network, or cannot afford the one-off connection fee.
Average prices paid by households for the different type of service have been
reported by David (2000); David, Inocencio, Clemente, Abracosa, Largo, Tabios,

and Walag (2000), based on surveys conducted in selected areas within the NCR
during 1995. More recently, Inocencio (2003) presented prices based on a survey
carried out in 2001. Table 9.2 summarises these prices and demonstrates the
higher price paid by communities that are not serviced by the MWSS network.

Note that in Table 9.2 a community water system is a more restricted definition
than that given by the NSO in Table 9.1. It is equivalent to a shared faucet whose

users pay a surcharge for the management of the service. Typically new housing
estates in outer suburbs are serviced in this manner, as the rate of MWSS network
expansion does not keep up with housing development.
For Manila in general, water expenditure takes a greater proportion of the
household budget than the rest of the country, and this proportion is highest for

149
 TABLE 9.2

Inocencio Study of Household Water Source

NOTE: This table is included on page 150 of the print copy of the
thesis held in the University of Adelaide Library.

($US1=P51)

150
lower income households. For the year 2000, the National Survey Office estimate
of annual household expenditure on water for households in the NCR was PhP

2,528 ($62) based on a population of 2,188,675 households. For the country ex-
cluding the NCR this falls to PhP 645 ($16) based on a population of 13,150,980
households. Although the budget share of water based on the same survey data
is quite low - less than one per cent - this figure is misleading because of large

variation in incomes. When the households are grouped into income deciles we
can see that for the NCR, the households in the top 10% of income pay only 0.7%
of their income on water while for the lowest 10% of earners the proportion of
income that is spent on water is 2.9%. For the rest of the country the expenditure

share for water from highest income decile to lowest is much lower - ranging from
0.7% to 0.2%. While households living outside the NCR may have access to more
alternatives to purchased water such as wells or water tanks, the distortions in
household water expenditure in the NCR compared to the rest of the country,
indicate that for many households water costs and pricing are an issue of major

importance. Table 9.3 shows the distribution of household expenditure on water

in NCR and for the rest of the country by income decile.

9.2 Consumption Analysis using Linear Methods

This section describes estimation of the domestic water demand function using
conventional linear models. We begin with a discussion of the process of prepara-

tion of a data set of household annual water consumption, expenditure and prices
that covers a sample of households in the NCR. This is followed by presentation
of econometric results and a discussion of results.

151
 TABLE 9.3

Waters Share of Household Budget (%)

Income Decile 1 2 3 4 5 6 7 8 9 10

NCR 2.9 1.8 1.5 1.9 1.9 1.8 1.6 1.3 1.2 0.7

Rest of Country 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.8 0.8 0.7

Source: calculated from NSO FIES 2000

9.2.1 Data Preparation and Description

The water data set is based on the 4,141 household observations within the
NCR from the FIES 2000 survey. For each observation record a weighted average

water price, unit value of bottled water, and household water consumption has
been determined. The use of unit values (expenditure divided by consumption)
as a proxy for price is discussed by Deaton (1997, p. 149). Water consumption is
the expenditure divided by the weighted average water price. A small number of

households (56) that have recorded no water expenditure are excluded from the
data set.
The weighted average water price was calculated by use of the NSO Housing
Census. Based on the given weights, the sample was converted to population
estimates. The variables related to source of water were then aggregated by city.

Proportions of households that pay for domestic water were calculated in three
categories: households with a water connection, those who share a connection,
and those using carted water2 . These form a set of weights that are used in the
average price calculation:
2
Households were excluded if their primary source is a well, river, lake or other free source.

152
 {wj,1, wj,2, wj,3}

where j is the city index, and the weights are ordered as direct, shared, and
vended.
Next two sets of prices were determined. The first set applies in the areas

serviced by MWCI and the second set in the areas serviced by MWSI. These
contain first, the all-in price (the average price per cubic metre charged by the
provider to which all taxes and levies are added); secondly the price of shared
connection, estimated at the all-in price plus 10%; thirdly, the price of vended

water which is assumed constant across the metropolitan area at PhP 50 per
cubic meter based on estimates provided by Inocencio (2003). The set of prices is
{pj,1, pj,2, pj,3}.
Finally, the weighted average price for city j is:

3
X
Pj = (wj,i pj,i)
i=1

The domestic water price estimates by city are shown in Table 9.4 below. The
summary statistics for the water consumption, pricing, and expenditure data set
for 4,085 households in the NCR are shown in Table 9.5.

153
 TABLE 9.4

Weighted Average Domestic Water Price (Pesos per m3 )

City Population Population Density Weighted Avg. Price

Manila 1,581,082 63,294 10.09

City of Mandaluyong 278,474 29,976 5.29
City of Marikina 391,170 18,177 7.06
City of Pasig 505,058 10,422 7.89
Quezon City 2,173,831 12,660 7.40
San Juan 117,680 19,778 4.71
Kalookan City 1,177,604 21,104 12.80
Malabon 338,855 21,569 9.08
Navotas 230,403 25,772 12.41
City of Valenzuela 485,433 10,324 18.35
City of Las Pinas 472,780 14,463 16.33
City of Makati 444,867 24,296 5.61
City of Muntinlupa 379,310 9,542 12.85
City of Paranaque 449,811 9,659 20.41
Pasay City 354,908 25,405 9.13
Pateros 57,407 5,520 5.82
Taguig 467,375 10,338 15.84
TOTAL 9,932,560 16,091

Source: NSO Population Census of 2000.

154
 TABLE 9.5

NCR Water Consumption, Pricing, and Expenditure Data Set:

Summary Statistics

Variable Description Mean Median Min. Max. SD.

QWATER water consumption (m3 ) 21.2 14.9 0.4 534.8 24.7

QBOTLE bottled water cons. (litres) 3.0 0.0 0.0 205.8 10.3

PWATER weighted average water price 10.4 10.4 4.7 20.4 4.3
155

PBOTLE bottled water price 18.7 18.7 10.0 30.3 1.9

WATER water expenditure 197.2 140.0 3.0 3,000.0 204.7

BOTLE bottled water expenditure 49.8 0.0 0.0 4,116.7 168.4

TOTEX total expenditure 18,819.8 13,849.3 884.8 515,791.7 21,654.4

FSIZE family size 5.0 5.0 1.0 19.5 2.2

(monthly consumption and expenditure)

(prices and expenditures are in Pesos)
9.2.2 Estimation of the Linear Demand Function

This section presents the results of estimation of the demand function using
two log-linear models and one non-linear model. In the first linear model the
coefficients are the estimates of price and income elasticity:

ln qi = 0 + 1 ln pi + 2 ln mi + 3 ln fi

where:

q = monthly HH consumption (m3 ) (QW AT ER)

p = weighted average price of water (pesos per m3 ) (P W AT ER)

m = monthly HH expenditure/income (pesos) (T OT EX)

f = family size (F SIZE)

The second is a log-linear model with an interaction term:

ln qi = 0 + 1 ln pi + 2 ln mi + 3 ln pi ln mi + 4 ln fi

where the variables are the same as for the log-linear model. This specification

appears in Rietveld, Rouwendal, and Zwart (1997). The elasticities are linear
combinations of the log of the other interaction variable:

price elasticity of demand: p = 1 + 3 (ln T OT EXi)

income elasticity of demand: m = 2 + 3 (ln P W AT ERi)

where is an aggregation function carried out over all observations.

The non-linear model alters the basic linear specification to include a power
term for the log price of water.

ln qi = 0 + 1 ln pi 2 + 3 ln mi + 4 ln fi

156
 TABLE 9.6

Estimates of Demand Function - NCR Water Consumption

model const ln p ln m ln f ln m ln p

basic -1.148 -0.5647 0.5034 0.1894

(0.183)* (0.0279)* (0.0176)* (0.0243)*
interaction -1.755 -0.291 0.5660 0.1901 -0.0283
(0.919) (0.406) (0.0945)* (0.0243)* (0.0420)

model const ln p expnt(ln p) ln m ln f

non-linear -2.740 3.55 -2.61 0.4836 0.1984

(0.330)* (1.55)* ( 1.39)* ( 0.0178)* (0.0244)*

NB: dependent variable: ln q

linear models: R2 = 0.284; = 0.71

This lets us model changes in price elasticity with water price itself:

p = 1 2 (ln P W AT ERi )2 1

This model has been estimated using non-linear least squares.

The estimation results are shown in Table 9.6. Note that in the basic linear

model all variables are significant, but in the model with the interaction term,
both the coefficients of price and the interaction term are not significant. This
suggests that model specification may be a problem. All coefficients in the linear
models have the expected signs. The non-linear model has significant coefficients

but the exponent term on log(P W AT ER) is significant only at the 90% level.

157
 TABLE 9.7

Price and Income Elasticities - NCR Water Consumption

Quartile

Model () Mean Min 2 Median 4 Max

Price Elasticity
basic linear -0.56
interaction term ln(T OT EX) -0.56 -0.48 -0.55 -0.56 -0.57 -0.66
non-linear ln(P W AT ER) -0.49 -1.90 -0.76 -0.53 -0.31 -0.17

Income Elasticity

basic linear 0.50

interaction term ln(P W AT ER) 0.50 0.52 0.51 0.50 0.49 0.48
non-linear 0.48

The elasticities are shown in Table 9.7. These are within the expected range,
have the expected signs and compare favourably with other studies (see for ex-
ample the meta-analysis conducted by Dalhuisen, Florax, de Groot, and Nijkamp
(2003)). Price elasticities in the interaction model are calculated at each income
quartile and the mean. For the non-linear model they depend on ln(P W AT ER)

and are calculated at each quartile and the mean. Note that in the interaction
model, price elasticity increases with budget. The reason for this is likely to be
that lower budget households have a lower discretionary use of water (for exam-
ple, car washing or use of washing machines) and are less elastic in their basic

water needs. The non-linear model gives us an insight into how prices themselves
affect price elasticity. According to the results, price increases when prices are
low have a greater effect on consumption that when prices are high. This result
is difficult to interpret because the model does not distinguish MWSS households

158
that pay a high marginal price because they consume more, from households who
rely on water from secondary sources and pay even higher prices. In the latter

case demand is likely to be less elastic because of basic needs. In the former case
demand at the higher rate may be less elastic because the household can easily
afford increased charges. In the case of income elasticity, there is little variation
in the figures - this suggests that income elasticity does not depend on the price,

and therefore does not depend on the source of purchased water.

In the following section we will estimate demand using a data set of households
that have a MWSS connection as their primary water source.

9.3 Consumption Analysis using Maximum Likelihood Methods

This section describes maximum likelihood estimation of the domestic water

demand function using the two error model described in Chapter 4. We begin with
a discussion of the process of preparation of a data set of household annual water

consumption. This is a selective data set in the sense that observations are limited
to those households with a connection to the MWSS supply. In contrast, the data
set used in the previous section was a sample of all households in Manila NCR
that pay for their water. Following from description of the data, we present and

discuss the econometric results of estimation of the two error demand function.

9.3.1 Data Preparation and Description

The NSO FIES survey data does not include information about the source of
domestic water for each household. To overcome this limitation, our approach is to
select two cities that have the highest proportion of households using the MWSS
network, based on the 2000 Household Census. The aggregated figures for NCR

were shown in Table 9.1. Disaggregating this data by city we find that the two

159
cities with the highest proportions of households with a direct or shared connection
are San Juan (99.41%) and Malabon (99.73%). These cities are situated entirely

within different service zones (MWCI and MWSI respectively) - and therefore
face different tariffs. This will increase the variance in prices, which we expect
will improve estimation results.

9.3.1.1 Calculation of Water Consumption

Discussion with staff at the NSO confirmed that during the FIES survey, when
households in connected areas were asked to estimate their water expenditure, they
simply presented the most recent invoice from the supplier. Therefore there is
justification for determining household water consumption by reverse application

of the suppliers bill. The domestic water bill is computed in the following manner.

1. Connection fee = PhP 9.25 (MWCI) 20.63 (MWSI)

2. Consumption charge = based on Table 9.8 below.
3. Basic Charge (A) = connection fee + consumption charge
4. CERA3 (B) = PhP 1 per m3 of consumption.
5. Water Charge (WC) = A+B
6. Environmental Charge (EC) = 10% of water charge
7. Maintenance Charge (MC) = PhP 1.50 (based on the meter type - residential
customers usually have a 13mm meter)
8. VAT = 10% of WC+EC+MC
9. TOTAL Bill = WC+EC+MC+VAT.

We now describe the algorithm to determine water consumption. For each

block, the marginal price is increased by one Peso to account for CERA. The
block base water charge is computed - this is the cost of consumption on the kink.
3
Currency Exchange Rate Adjustment.

160
 TABLE 9.8

Tariffs for Domestic Water Use - Manila 2000

cubic metres
Firm fixed 10-20 20-40 40-60 60-80 80-100 100-150 150-200 200+

MWCI 9.25 1.13 2.14 2.82 3.29 3.45 3.60 3.76 3.91

MWSI 20.63 2.51 4.79 6.29 7.35 7.68 8.03 8.39 8.74

Note: fixed charge connection fee includes usage up to 10m3 .

Block prices are pesos/m3 .

For example, the MWCI block 1 base is the PhP 9.25 connection fee, while the

block 2 base is PhP 9.25 + PhP 10 (10 m3 at a marginal rate of one Peso) =
PhP 19.25, and so on. Block bases define the range of expenditure for each block,
so each households water charge must map to a unique block. The consumption
for a household is computed by removing the VAT, MC, and EC from the bill,

leaving the water charge. The block is determined, and consumption is the sum
of the block base consumption and the block marginal consumption.
In creating the consumption data set, we have selected households in those
cities where the primary source of water is the MWSS network. This does not

preclude supplementation by other sources of water, or the small (less than 0.6%)
probability that the household does not have a connection to the network. In the
latter case, we assume that household expenditure is higher than an equivalent
household with a connection and so the estimated consumption would also be
higher. In the former case, when MWSS water is supplemented with free water

(there would be no reason for a household to purchase more expensive water), ex-
penditure would be lower than for an equivalent household without a free source,

161
therefore the estimated consumption would also be lower. We also need to con-
sider the effect of including households with a shared connection in the data set.

Such households may pay an higher or lower average price depending on whether
the owner of the connection applies a surcharge and the marginal rate of con-
sumption. Groups of households with low total consumption can expect to pay a
lower average price because the connection fee is shared; however at higher levels

of consumption the average price would increase above that paid by a household
with a direct connection4
Given the above, there appears to be justification for assuming that the errors
in our estimate of water consumption that are due to supplementation of water

sources, inclusion of non-MWSS users, and connection sharing, will be random

normal with zero mean. These errors are in addition to the errors generated
within the survey process itself.

4
Sharing is usually among 2-3 households.

162
 TABLE 9.9

MWSS Water Consumption Data Set : Summary Statistics

Variable Description Mean Median Min. Max. SD.

QWATER water consumption (m3 ) 34.78 29.82 1.78 130.43 21.61

WATER water expenditure 144.05 100.00 15.00 800.00 114.16
163

TOTEX total expenditure 17,077.00 11,964.00 2,296.00 133,533.00 17,181.65

FSIZE family size 4.79 5.00 1.00 13.00 2.24

CITY city code

Notes:
- monthly consumption and expenditure
- prices and expenditures are in Pesos
- cities are Malabon (MWSI) and San Juan (MWCI)
- sample size 355 households in population of 98,742 households
9.3.1.2 Descriptive Statistics

Table 9.9 shows the descriptive statistics for the MWSS consumption data
set. The mean value of household consumption compares favourably with other
sources. MWCI reported residential account holder consumption of 41,828,150 m3

for the three months from January to March 2001. They also estimate that there
are 413,242 households serviced by their network (including those with shared
connections and those that participate in Tubig para sa Barangay programmes.
Therefore the estimate of mean consumption for MWCI serviced areas is 33.7 m3 .

In that same period MWSI delivered 73,633.05 m3 to customers across 602,424

households, so the mean consumption is 40.7 m3 . The ADB estimates (McIntosh
and Yniquez, 1997) of mean consumption for households with a MWSS connection
is 44 m3 . Other collaborating estimates of mean household water consumption

for MWSS customers are 34.7 m3 (David and Inocencio, 2002). The same authors
also report on average consumption ranging from 6.8 to 11.4 m3 for households
that do not have a MWSS connection and who rely on the secondary market. This
explains the much lower average consumption (21.2 m3 ) that appears in Table 9.5
that is based on the weighted average water price and includes all households that

pay for their water.

9.3.2 Estimation Results

The two error model was estimated using maximum likelihood technique with
the likelihood function as presented in Section 4.3.2.2 modified to cover a nine
block tariff. Again, we have used a log-linear model with interaction term where

the elasticities are linear functions of the other interaction variable:

ln qi = 0 + 1 ln pi + 2 ln mi + 3 ln pi ln mi + 4 ln fi

164
 The maximum likelihood estimates have been estimated by use of a computer
programme written in the R statistical processing language (R Development Core

Team, 2006). As each supplier has a different tariff (but the same block struc-
ture), the programme applies different prices according to the CIT Y variable.
Maximisation of the log likelihood function is performed using the algorithm of
Nelder and Mead (1965) which, although relatively slow, is quite robust and is

suitable for highly non-linear and non-differentiable functions. Comparable re-

sults were also achieved using an optional conjugate gradients method of Fletcher
and Reeves (Fletcher and Reeves, 1964). Standard errors of the parameters have
been computed by taking the square root of the diagonal of the inverted Hessian

matrix, and these have been used to compute t-statistics.

The estimated coefficients and the elasticities are shown in Tables 9.10 and 9.11
below. The standard errors and t-statistics indicate that all variables but income
are significant. However, to confirm this, a likelihood ratio test has been carried
out. This involves estimating a restricted and unrestricted model and comparing

the ratio of likelihoods with a chi-squared distribution. The null hypothesis is

that the two models are the same and therefore the coefficient is not different
from zero. The results of this test are shown in table 9.12 and, as we reject the
null hypothesis, we conclude that income does have a non-zero coefficient.

On first sight the estimates of the standard deviation of the two error distri-
butions appear quite large. Indeed considering the nature of the data, this should
not be surprising. However, these are not in log units and equate to cubic metres
of consumption. Therefore we can see that the error term has a standard de-

viation of just over the average block width of 25m3 while the error term has a
standard deviation of about one quarter that. The model was designed specially
to cater for household consumption in another block to the optimal choice, and in
this case, appears to have accounted for that situation quite well. Exposure to a

165
variety of additional data sets would increase our knowledge as to the conditions
under which heterogeneity affects household consumption choices.

The elasticities are of considerable interest. First, as there is an interaction

term, the price elasticity varies with the budget. The results clearly show that
as incomes increase, household consumption of water becomes less price elastic.
However, closer examination shows that (because of the distribution of incomes)

elasticity is fairly constant over the range of incomes, but drops off significantly in
the higher quantiles of income. This indicates that higher income households are
much less likely to reduce consumption in the face of increased prices than median
and lower income households. Note that this result stands in contrast to the result

shown in Table 9.7 - where we suggested that lower income households were less
willing to decrease consumption than higher income households. Comparing the
two, households with a connection to the MWSS network have a substantially more
elastic response than households in general. The reason for this most likely to be
that they are consuming more in the first place (refer to the mean consumption

figures in Tables 9.5 and 9.9); it is easier to cut back on nonessential uses in
the face of price increases. Finally, income elasticities show an increase as prices
increase. This means that higher income households will consume more water
as their incomes increase and that this effect increases with higher consumption

(as the block rate is increasing). Although more research is needed, these results
suggest that price and income elasticities are an important factor in block rate
tariff design. This type of information can be used by regulators and utilities
to predict the likely response to price changes for consumers grouped by current

consumption and income - thereby helping to achieve specific revenue outcomes

and to manage the resource. Cross subsidisation may also be an objective of the
tariff design where knowledge of price elasticity for the different groups is essential.
Households who consume more water per capita than average would be required

166
 TABLE 9.10

Estimates of Demand Function - MWSS Connected Households

Model const ln p ln m ln f ln m ln p

interaction 5.41 -4.72 -0.050 0.2572 0.354 26.430 5.587

(2.45)* (1.77)* (0.245) (0.0815)* (0.174)* (1.801)* (0.819)*

dependent variable: ln q
log likelihood: -1478.86

to pay above the marginal cost of supply while those who consume less per capita
would pay less than the marginal cost.

167
 TABLE 9.11

Price and Income Elasticities - MWSS Connected Households

Quartile
Model () Mean Min 2 Median 4 Max

Price Elasticity

interaction ln(T OT EX) -1.27 -1.98 -1.52 -1.40 -1.25 -0.54

Income Elasticity

interaction (MWCI) ln(P W AT ER) 0.41 -0.05 0.36 0.47 0.49 0.51

interaction (MWSI) ln(P W AT ER) 0.64 -0.05 0.57 0.70 0.73 0.76

TABLE 9.12

Likelihood Ratio Test

Null Hypothesis: ln(T OT EX) is not a significant determinant of consumption.

H0 : 2 = 0 H1 : 2 6= 0 DF: 1 2 95% critical value: 3.841

L(U R ) L(R ) Test Statistic Decision

-1478.86 -1642.88 328.04 reject H0

nb: = 2 (L(R ) L(U R ))

168
 CHAPTER 10

Second Case Study

Urban Water Services in Manila
Part C - Marginal Cost and Welfare

10.1 Introduction

Part A of this case study presented estimation results of the constant returns
to scale and variable returns to scale cost functions for the Manila water conces-
sions. We observed a significant improvement in model fit for the variable returns
compared to the constant returns models. The results indicated that the conces-

sions exhibited decreasing returns to scale in production, however the hypothesis

of constant returns to scale was not rejected in favour of the alternative of de-
creasing returns. The estimated returns to scale parameter was 0.767. Part B
of this study estimated demand for water using a metropolitan wide sample that

contained households that pay for water from a variety of sources; and a restricted
sample that, with few exceptions, consisted only of households with a connection
to the MWSS network. The price elasticity of demand was estimated to be -0.56
for the former sample and -1.27 for the latter sample. We postulated that the

difference in elasticities between the two samples was related to the former having
a lower mean consumption and therefore having a higher proportion of their water
for essential needs; in contrast to the connected group sample that consumed more
and had a larger proportion of water available for non-essential use.

169
 In this part of the case study we turn our attention to issues of welfare in the
pricing of urban water services. We will restrict ourselves to consideration of the

welfare effects for consumers who have a piped water connection. The reason for
this is that policy can address prices set by the regulator for this form of service,
whereas local independent entrepreneurs who supply the remaining households
operate in an unregulated market outside the control of government.

We commence by presenting variable return to scale cost functions evaluated

at the mean values of co-variates and at a partial equilibrium of marginal cost and
benefit functions. This will be followed by estimation of several welfare statistics
that compare current price/demand values with the efficient values obtained at

the partial equilibrium. We conclude with some overall observations related to

this case study.

10.2 Marginal Cost Estimation

In this section we calculate the marginal cost of a household water connection

in KL per month. This will provide a reference to compare with the current
monthly tariff. We will base this on the estimated cost and demand functions for
the Manila water concessions and employ three methods: marginal system cost

at mean output, marginal cost at output per connection, and marginal cost at an
equilibrium of supply and demand. Table 10.2 shows sample means for parameters
used in calculations. We have used the combined means rather than the values
disaggregated by concessionaire. Prices are the average tariffs set at January 2000
including the currency exchange rate adjustment (CERA) but excluding all taxes.

The results are summarised in Table 10.1.

170
10.2.1 Marginal System Cost

The annual cost function was shown in Table 8.5. It exhibits decreasing returns
to scale Cobb-Douglas production:

C = 3.349 r 0.266 w 0.734 x1.317

This function was estimated with output in MLD, and cost and factor prices
in K. Pesos. Differentiating with respect to output,

MC = 424.836x0.317

The annual marginal cost in PhP of a one MLD increase at mean output (MLD)

of 740.245 is estimated to be PhP 3,449.958. Converting this into a monthly cost

per KL gives a marginal cost of PhP 9.45.

10.2.2 Marginal Cost in Per Connection Terms

This approach alters the units of the cost function so that marginal cost is
determined in per connection terms. Note this is still the marginal cost of an

increase in output, not the marginal cost of an additional connection.

We express annual cost as a monthly per connection cost with output in m3
per month:

 1.317
1 1000 12
h
C = 3.349 r 0.266 w 0.734 (H x)1.317
H 12 365000

where C h is monthly per connection cost in Pesos, H is number of connections,

and x is per connection output.

At the factor price sample means of r = 5.45 and w = 272.243, with H =

887, 371, the decreasing returns to scale cost function evaluates to:

171
 C h = 2.583 x1.317

The marginal cost is therefore:

MC h = 3.402 x0.317 (10.1)

At the mean monthly household consumption of 32.78 KL, this equates to PhP
10.25.

10.2.3 Marginal Cost at a Partial Equilibrium

This third method uses a partial equilibrium of the supply and demand curves

to determine marginal cost at the efficient level of output. We use the marginal
cost function in monthly per connection terms derived in the preceding section
and the household demand function that was derived by use of the two error model
for a block rate tariff.

We need to point out that the application of partial equilibrium analysis to

determine marginal cost is one way in which a block rate tariff can be designed.
For example, individual demand curves for groups of consumers with similar con-
sumption levels would be combined with the marginal cost data to establish the

marginal cost for a range of output1 . By increasing the marginal rates for house-
holds with high levels of consumption, the utility can cross-subsidise households
with low levels of consumption and ensure that their marginal utility of expendi-
ture is higher. As this approach requires a detailed level of cost and consumption

data that is beyond the reach of this study, the partial equilibrium analysis that
follows assumes only a single marginal rate and one consumer group with no
cross-subsidisation.
1
The industry approach is to use present value estimates of average cost and demand.

172
 TABLE 10.1

Summary of Marginal Cost Estimates: Manila Water Concessions

Marginal Cost

Evaluation Method Output (PhP/KLM) Charge(**)

Variable Returns to Scale

system mean output 740.25 MLD 9.40 5.84
per connection mean output/cons. 32.78 KLM 10.20 5.84
partial equilibrium 14.81 KLM 7.93 5.84
** average price of both concessionaires

The parameter estimates of household demand function with an interaction

term (from Table 9.10) were:

ln x = 5.41 + ln p(4.72 + 0.354 ln m) 0.050 ln m + 0.257 ln f

Substituting mean values for monthly income m = PhP17, 077 and family size
f = 4.79, and removing logs, we have the household monthly demand function at

the means:

x = 205.54 p1.27 (10.2)

Solving equations 10.1 and 10.2 for x and p yields the equilibrium price and
consumption level: p = 7.93, and x = 14.81.

173
10.3 Welfare Effects at Average Price Levels

If we refer to Table 10.2, is is apparent that average prices are less than the
marginal cost estimates obtained in the previous section. In Chapter 5 we pre-
sented the methodology for calculating the deadweight loss arising from pricing
below marginal cost based on Renzetti (1999). In this section we calculate the

deadweight loss based on these estimates of marginal cost and the prevailing av-
erage prices corresponding to the data set. Current consumption x0 is determined
from the demand curve at the prevailing average price.
The welfare estimates appear in Table 10.3. This table includes sensitivity

estimates of two critical parameters: the intercept of the cost function, and the
coefficient of output (cost elasticity). Each of these parameters have been varied
by 10% of their standard errors. Figure 10.1 shows the marginal cost and bene-
fit curves with an overlay of the 2000 average price and predicted consumption.
We can see from this diagram that the the equilibrium price is higher than the

weighted average price, and that this results in a loss in welfare. The formulae for
the welfare statistics are repeated below:

deadweight loss: loss of welfare at current prices

R x0
DWL(x0 ) = x
(MC (x) MB (x)) dx

deviation: deviation from efficient output

DEV (x0 ) = (x0 x ) /x0

waste: average DWL per unit of current output

waste(x0 ) = DWL(x0 )/x0

174
 TABLE 10.2

Sample Means of Parameters: Manila Water Concessionaires

Factor Variable Sample MWCI MWSI

monthly income (PhP) m 17,077 21,906 12,220

family size f 4.79 4.74 4.84
cost of capital (%) r 5.45 5.15 5.76
annual wage rates (PhP 000s) w 272.243 272.266 272.221
connections H 887,371 339,491 547,880
mthly per connection cons./output x 32.78 43.17 26.34
3
price (PhP/m ) p 5.84 3.76 7.13

TABLE 10.3

Welfare Measurements: Variable Returns Model

minus 10% SD. plus 10% SD.

1 1
parameter units Mean Est. 0 Q
0 Q

avg. price Jan. 2000 PhP/KL 5.84

qty. demanded 2000 KLM 21.84
efficient price PhP/KL 7.93 6.19 5.98 10.16 10.23
efficient output KLM 14.81 20.29 10.72 10.81 21.20
DWL per connection PhP 12.00 0.39 0.06 43.87 48.00
total DWL KPhP 10,656 346 53 38,928 42,581
deviation pct 32.20 7.10 2.95 50.52 50.91
waste PhP/KL 0.55 0.02 0.00 2.00 2.20

175
 Figure 10.1. Partial Equilibrium for the Manila Concessions

10.4 Conclusion

Based on a sample of data from 1997 to 2005, there is evidence that the produc-
tion of urban water in Manila exhibits decreasing returns to scale and increasing
marginal costs2 . Demand, based on data from a household expenditure survey
is substantially price elastic but at higher levels of consumption for households

with a water connection. This is in contrast to the less elastic price response
for households who lack a connection and rely on more expensive water supplies.
Estimates of marginal cost evaluated at sample data mean values of consumption,
and at a partial equilibrium of marginal cost and benefit curves indicate that, in

the year 2000, the average price of MWSS water was less than its marginal cost.
This suggests that a piped water connection delivered benefits in excess of its
2
The hypothesis of constant returns to scale was not rejected however.

176
cost to consumers, resulting in higher consumption for those households with a
water connection in comparison to those without one. Estimates of annual system

deadweight loss caused by inefficient pricing are approximately PhP 10.7 million
(US$0.24 million). The analysis indicates that a price increase of about 36% would
be required to reach the efficient level of output. Because demand would decline
with a higher price, a better approach might be to increase by 6% (the lower

estimate of efficient price) and monitor changes to demand. Sensitivity analysis

indicates that adjustments to coefficients can result in substantial changes to the
results - this should not be surprising as the coefficients are variable exponents in
the cost equation.

Substantial tariff increases were awarded in 2001 and 2003 to both MWSS
concessionaires. In January 2003, at the beginning of the new five year rate
rebasing period, the all-in price for MWCI was increased to PhP 13.43/m3 while
for MWSI the new average price became PhP 19.92/m3 . These are both well above
the efficient price estimated here, even when adjusted for inflation. The basis of

non-CPI price increases awarded to the concessionaires are primarily the increase
in foreign debt service costs brought about by the currency devaluation during
the Asian Economic Crisis of late 1997. These increased costs would be reflected
in increases in the cost of capital and therefore we would expect the marginal cost

estimates to increase once the variable returns model was re-estimated. Prices in
Manila are regulated under a rate of return system, which, as noted in Chapter
2, has substantial reporting requirements in comparison to price cap regulation
and can involve lengthy negotiations with the regulator3 . This implies that price

adjustments themselves add to compliance overhead costs under rate of return

regulation.

3
In Manila it took four years to award price increases based on currency devaluation after
the 1997 Asian Economic Crisis.

177
 CHAPTER 11

Performance Based Pricing

11.1 Introduction

In Chapter 2, we briefly noted some of the limitations of the marginal cost ap-

proach to water pricing. This chapter expands on these, based on the experience
gained in the case studies. In particular we present an alternative to the main
behavioural assumption implicit in empirical analysis - that utilities strive to min-
imise costs, because in the regulated environment this is the only way that profits
can be increased. This alternative approach views firms as heterogeneous in their

efficiency, with some able to perform at higher levels than others based on a set
of benchmarks. We posit that the cost functions of the better performing utilities
be used as the basis for marginal cost pricing methods described in previous chap-
ters. Performance based pricing is likely to become important as competition is

introduced in urban water supply and changes in the regulatory environment see
the introduction of benchmarking and performance based regulation.
In this chapter we first review some of the main findings of the two case studies
that were presented in Chapters 6-10. We then examine some issues related to

the marginal cost approach and question the assumption that firms focus on cost
minimisation. This chapter then presents a method for estimation of efficiency
that has been used in benchmarking studies - Stochastic Frontier Analysis. In
the chapter that follows, we will present a case study in which this technique will

178
be applied to a cross section of water utilities, and discuss some consequences of
performance based pricing.

11.2 Review of Case Studies

The case studies contained in the previous five chapters have applied marginal

cost principles to water utilities in Victoria and the Philippines. We observed in

both cases, based on empirical analysis of sample cost and demand data, that
average prices prevailing at the time were substantially less than the estimated
marginal cost of supply. For the Victorian water businesses, estimates of the dead
weight loss at current prices were 3.3% of revenues for metropolitan and 7.2% of

revenues for the regional businesses. For the Manila concessions the dead weight
loss as a proportion of revenue was estimated at 9.4%.
For the Victorian water businesses recovery of these losses primarily occurs
through two forms of redistribution. First, metropolitan water business have

generally been able to return accounting profits to the state government, part of
this is redistributed to regional businesses - who are not profitable - in the form
of government contributions. Second, income is received from connection fees,
and new distribution assets are handed over to the utility by developers of new

housing areas at city fringes1 . In the case of the Manila concessions losses are
born entirely by the owners. Because the government has an interest in ensuring
that concessionaires remain financially viable, and if these costs are unavoidable,
prices must be adjusted so that costs are recovered.
We saw evidence of increasing returns to scale (decreasing marginal cost) for

the Victorian water businesses and decreasing returns to scale (increasing marginal
cost) for the Manila concessions (although the null of constant returns could not
be rejected). Application of marginal cost pricing in Victoria would result in
1
These are termed Gifted Assets.

179
efficient levels of consumption but subsidies or a form of Ramsey pricing would
still be required to recover costs. With increasing returns the quantity of water

demanded will be below the predicted level of demand when either average or
marginal cost pricing is used - providing a safe buffer of capacity2 . For the Manila
concessions, or any utility that operates under decreasing returns to scale, there
is the risk of erosion of this safe buffer when average cost pricing is used and

predicted demand is underestimated. This is because the low price encourages

consumption in excess of what has been costed for. This risk also exists under
marginal cost pricing and decreasing returns to scale, but is less because prices
are higher.

In general, factor price cost elasticities3 for the Victorian water businesses
are higher for labour than capital, but when depreciation is included as a cost,
this is reversed. We expect that as more capital investment is undertaken by
the Victorian water businesses, overall expenditure will be more responsive to
the cost of capital. For the Manila concessions, the difference in elasticities is

more apparent. The cost elasticity for labour is significantly greater than for
capital. Again we would expect that as investment increases, capital cost shares
also increase.
The case studies represent contrasting tariff structures. In the case of Victorian

water authorities for the period of analysis, tariffs were predominantly based on
a single volumetric rate with a basic access charge. For Manila an increasing nine
block rate tariff is used. Modelling demand under block rate tariffs is complex as
the demand curve is non-linear. The estimates of price elasticity for consumers

serviced by the Manila concessions are comparable with other studies, for example
in Indonesia (Rietveld, Rouwendal, and Zwart, 1997), but are substantially higher
2
The marginal cost price must be at or above the equilibrium price.
3
These are equal to cost shares in the CES models.

180
in comparison to a sample consisting of all households including those without a
connection. This supports the stylised fact that lower income households, those

who cannot afford a connection or who live in areas not serviced by the network,
have a less elastic response to increases in the water price, in comparison to those
with a permanent water connection.
The case studies are contrasted in the form of utility ownership and in the

regulatory environment. The Victorian water businesses are fully owned by state
government. They are regulated by the states Essential Services Commission
(ESC). Prices are set by each business but must be agreed by the ESC and in
accordance with business existing Water Plans. The basis of price is average

cost using discounted historical and projected costs and demand. The Manila
concessions are private concerns under which government retains ownership of
existing assets and assumes ownership of new assets at the end of the concession
period of 25 years. Prices are set by rate of return regulation where the regulator
decides allowable costs and the return on capital. In the event that prices were set

at marginal cost we expect the outcome in each case to differ. For the Victorian
businesses, prices would remain below average cost and therefore subsidisation
would still be required. For the Manila concessions, with marginal cost increasing,
there is a possibility of windfall profits. The concessionaires would need to reinvest

these profits to avoid the regulator forcing a reduction in price.

The marginal cost approach is based on economic principles of efficiency and
equilibrium between supply and demand for the service. In this study we have
employed statistical approaches based on a number of cost and demand models.

This contrasts with industry approaches that are primarily based on present value
of historical and forward predictions of average cost. Both approaches have con-
siderable data requirements but the average cost method necessitates use of a
discount rate. We have seen that a functional approach is dependent on knowl-

181
edge of demand and factor prices, our approach uses mean values and a welfare
maximising partial equilibrium to arrive at the marginal cost. A hybrid approach

is also possible where marginal cost is determined by evaluation of the marginal

cost function at the present value of projected annual demands.
There will be greater focus on efficiency and marginal cost approaches in the
future, caused by changes in the regulatory environment, increased competition in

the water sector, and awareness of scarcity of the resource. As noted in Chapter
2, no single solution or formula exists in water management and pricing. Regu-
latory authorities and utilities will need to utilise an array of tools and evaluate
the merits of each alternative outcome. In the following sections we reconsider

the problem of pricing based on marginal cost to establish the motivation for
alternative performance based and hybrid approaches.

11.3 Price and Sustainable Water Management

An increase in the price of urban water can be expected to have three main

effects. First, the full cost of service provision is more likely to be met. Second,
long term funding commitments can be made for investment in infrastructure or
technology that will ensure the sustainability and quality of future supply. Third,
price increases can reduce water consumption to adjust for reduced inflows caused

by seasonal or long term climatic conditions.

Despite some signs of change, governments have generally been reluctant to
use price to achieve these objectives. This is for several possible reasons: water is
an essential service and governments are keenly aware that price increases may be

politically risky; price increases may place some people under financial hardship
and require introduction of rebates or similar measures that have a cost; high water
users may not respond to increased prices; and windfall revenue increases through
higher prices may also be politically unpopular. Water restrictions, backed up

182
by monetary penalties are used by Australian state governments in preference to
price increases to cope with increased seasonal demand or reduced supply. This

is despite there being no evidence that consumers would be unable to adjust to

fluctuating water prices. As we have seen in Manila, private owners will increase
prices so that the full cost of provision is met, subject to the constraints of the
regulator.

Price is not the only lever that can be used to ensure sustainable water sup-
plies. Public awareness campaigns can have an impact, for example the Water
Proofing Adelaide programme (Government of South Australia, 2005). There are
numerous options for improvement of urban water management: stormwater re-

cycling, greywater, and wastewater recycling, and desalination are all feasible. In
Australia, the federal government has initiated a programme involving repurchase
of agricultural water entitlements, while state governments are purchasing water
allocations from agricultural producers.
In the case studies we examined the welfare effects of current pricing. The

measurement of welfare by present value of net benefit is a basic tool of cost-benefit

analysis that has been in use for over two decades. When there are externalities,
however, a purely monetary measure of well-being presents some shortcomings.
Consider the case where a water utility invests in water quality technology. The

marginal cost will increase and, unless the investment is subsidised, the price
must eventually adjust. When this occurs either consumers demand less of the
more expensive water, or they are willing to pay more for the same quantity of
better quality water. In the first case, there is a decrease in consumer surplus,

while in the second case the demand curve shifts outward and consumer surplus is
unchanged. We expect that there is, however, a clear benefit of improved health
to those households who consume the better quality water. The externality will
only be captured in the cost benefit analysis if consumption also increases because

183
the water quality has improved.

11.4 Cost Minimisation in the Supply of Urban Water

In this study, we have taken the view that price and output are exogenous
to the firms cost model. Practically the only option that a regulated utility
has to increase profits is by cost minimisation. Under some conditions however,

the behavioural assumption of cost minimisation might be flawed. First, for a

rate of return regulated utility, it might be sufficient to hold costs at the level
agreed by the regulator so that the allowed return on capital is achieved, thereby
avoiding the possibility of windfall profits. The regulatory environment provides a

disincentive to stray from the agreed cost structure by increasing efficiencies, and
investors will not demand cost reductions providing returns on their investment
match expectations. Second, cost minimisation is generally applied to competitive
rather than regulated markets. The regulator may try to make up for the absence
of competition but this does not guarantee that firms will behave as they would in

a competitive environment. Firms may not operate in markets where factors are
traded competitively. Capital may be only available at higher than market rates
because the lending risk is perceived higher, suppliers may overprice their services
because they regard the utility to be a monopoly, or alternately the utility may be

able to exercise monopsony power in its purchase of labour and services. Third,
firms may be prepared to undergo short term cost inefficiencies so that they receive
long term efficiency gains. An investment decision, for example, may be guided
by this approach. Despite these potential flaws, the behavioural assumption of

cost minimisation dominates most applied research in the area of cost modelling
of utilities.
In the case studies we estimated constant elasticity of substitution production
and cost models. The results can be used to predict expected or average production

184
and cost for a representative utility at different input levels or factor prices. If
we relax the assumption that firms minimise costs then a minimum cost function

will not explain the sample as well as expected. The distributional assumptions
of the models may mask individual differences in firm behaviour; in particular,
those that effect cost efficiency. We may be unable to account for these a-priori
as we do not know what variables to measure. We can however, introduce error

terms to the model to account for more heterogeneity in firm behaviour as we did
for households in the block rate demand model presented in Chapter 4. Pricing
based on assumptions about cost minimisation may yield incorrect results if the
marginal cost has been determined from a sample that contains inefficient firms.

This may lead to consumers paying higher than necessary prices.

11.5 A Performance Based Approach to Urban Water Pricing

The foregoing discussion suggests several initial objectives of a performance

based approach to water pricing. First, it should model a frontier of performance,

rather than mean cost minimising behaviour. Second, there should be a metric for
the distance of each firm from the frontier. Third, although production and cost
are the usual criteria for measuring performance, other measures of efficiency, for
example service and water quality, could be incorporated into a multiple output

model.
The measurement of performance and efficiency has received considerable at-
tention in the literature. In Saal and Parker (2004) the authors use a measure of
Total Factor Productivity (TFP) to determine efficiency under a RPI+K regula-

tory pricing system for a sample of English and Welsh water utilities. Fox and
Hofler (1986) used composed error frontiers to measure water utility efficiency
with a dual output model and a sample of US rural water producers. Tupper and
Resende (2004) used a form of Data Envelopment Analysis (DEA) to examine the

185
efficiency of water and sewerage companies in Brazil. This work has been followed
by two studies, also based in Brazil, that use Stochastic Frontier Analysis (SFA):

Sabbioni (2005), and Faria, da Silva Souza, and Moreira (2005). Estache and
Rossi (2002) use Stochastic Frontier Analysis to measure the efficiency of water
utilities in the Asian Pacific region. They use error components and technical effi-
ciency effects as measures of efficiency. While much of the applied work has been

dominated by interest in the public/private ownership debate; recently, pricing

of utilities based on performance based measures has received attention, particu-
larly to establish the X-factor for RPI+X price cap regulation (Thakur and Singh,
2005).

Stochastic Frontier Analysis (Aigner, Lovell, and Schmidt, 1977; Battese and
Coelli, 1988; Jondrow, Lovell, Materov, and Schmidt, 1982), and (Greene, 2003,
p. 505) rests on the notion that firms in practice do not achieve the theoretical
optimum levels (frontiers) of output maximisation or cost minimisation in produc-
tion. Instead, sub-optimal levels of output or cost are achieved and the proximity

to the optimum is governed by firm level heterogeneity. Furthermore this prox-

imity is measurable and can be considered to be a measure of individual firms
efficiency.
SFA contrasts in two ways with the measurement of the welfare maximising

level of output described in Chapter 5 and applied in the preceding two case
studies. First, the parameter of interest is the efficiency distribution rather than
the parameters of the production/cost function itself. Second, it provides a ready
measure for comparison of firms with similar characteristics. Indeed it is possible

to rank firms according to their SFA measure of efficiency. In the preceding two
case studies we showed that there were significant deadweight losses involved in
current prices4 , therefore welfare would be increased if prices were increased. To
4
Based on the time of the data set.

186
contrast this, the SFA approach is that welfare is increased by benchmarking -
where firms improve their performance by striving to reach efficient frontiers set

by and among their domestic or international peers. One other distinction, not
addressed in the literature, is that frontier methods are focused on firm behaviour
and do not directly include consumer demand in the efficiency model.

11.5.1 Stochastic Frontier Analysis

The stochastic frontier model augments the basic linear regression model with

a non-negative random term that is a firm specific measure of sub-optimality or

distance from the efficient optimum:

yi = xi + vi ui

where x is a vector of explanatory variables for the ith cross sectional unit,
and v is an assumed random disturbance due to idiosyncratic firm specific effects
(vi N[0, v2 ]), uncorrelated with any other variables. The inefficiency term ui is

non-negative random term, which, when data is in log form is a measure of the
percentage by which an observation fails to reach the frontier.
Much of the theoretical work in SFA has focused on estimation of SFAs with
different distributions of u. Two of the most common are the absolute value of a

normal distribution (the half normal) and an exponentially distributed variable.

The log likelihood of the half normal is:

n n  
n 2 1 X  i 2 X i
ln L = n ln ln + ln (11.1)
2 2 i=1 i=1

where 2 = u2 + v2 , i = vi ui, = u /v , and is the standard normal

probability distribution.
Although numerical methods may be used to maximise this function, an an-

187
alytic solution is available by a process of substitution, and deriving the second
differentials. The parameter estimates are then obtained by back-substitution into

the asymptotic covariance matrix. This process is described in detail in Greene

(2003, p. 504) and will not be reproduced here.
As mentioned the parameter of interest in SFA is usually u rather than the pa-
rameters of the production or cost function. The procedure above only provides

a firm level estimate of the total error . An approximation that is in widespread

use is attributed to Jondrow, Lovell, Materov, and Schmidt (1982) (for the half
normal model):

 
(z)
E[u|] = z
1 + 2 1 (z)

z=

and, for the exponential model:

 
(z)
E[u|] = z
1 + 2 1 (z)

z=

The proportion of the total variance of that is due to the inefficiency term
is half normal:

var[] = var[u] + var[v] = (1 2/)u2 + v2

and for the exponential model this is 1/2 + v2 .

188
 CHAPTER 12

Third Case Study

Performance Based Pricing:

A Frontier Analysis of Asian and Australian Water Utilities

12.1 Introduction

The objective of this case study is to present the results of estimation of a cost

efficiency frontier for a cross section of water utilities, using Stochastic Frontier
Analysis as the econometric tool. We use a sample of data obtained from a World
Bank sponsored utility benchmarking database to estimate the frontier for two
log-linear cost models, and present the cost efficiency estimates of the utilities in
the sample. We propose that a regulator could set the volumetric price at the

marginal cost of the most cost efficient firms1 , based on a ranking or grouping of
utilities by efficiency. Alternately, for a price cap form of regulation, data from
the most efficient firms can be used to set the X-factor. We have not estimated
X-factors or marginal cost for the firms in this sample as this requires data on

costs and production from each utility at the same level of detail of the first and
second case studies.

12.2 Source of Data

The input data set originates from public domain data available at the Inter-
national Benchmarking Network for Water and Sanitation Utilities (The World
1
Reasonably we would expect the sample to be restricted to one country in this case.

189
Bank, 2005). This is an initiative of the World Bank that started in the late 1990s
and is supported by the UK Government Department for International Develop-

ment (DFID) and the Banks Water and Sanitation Programme. The IBNET
aims to support and promote benchmarking in the global water and sanitation
sector by providing data, software and documentation to assist utility managers
to undertake benchmarking and analyse their own and their peers performance

information over time.

Benchmarking, in the IBNET context, is the comparison of service perfor-
mance among peer utilities - the goal being to enable improvement in quality and
performance. The motivation for using benchmarking is well expressed in their

own words:

If benchmarking contributes to improving utility performance, cus-

tomers will benefit when the savings result in improved service and/or
lower tariffs. Policymakers and regulators can compare sector perfor-
mance with that achieved within the country and with other countries,
while global institutions (such as the international finance institutions)
can improve policy advice based on using available sector performance
data.
. . . Inter-utility performance comparison is needed in the water and
sanitation sector, because the sector offers limited scope for direct com-
petition. Firms operating in competitive markets are under constant
pressure to out perform each other. Water utilities are often sheltered
from this pressure, and it frequently shows: some utilities are on a
sustained improvement track, but many others keep falling further be-
hind best practice. This matters, because a well-run water utility is
essential to peoples lives. Only the most efficient, financially viable
utilities are able to respond to urban growth, connect the poor, and
improve wastewater disposal practices. (source: www.ib-net.org)

A number of countries and regions have initiated their own performance bench-

marking in addition to contributing to IBNET. These include the UK Office of

Water (OFWAT) which benchmarks against utilities in Australia, the Nether-
lands, Canada and the United States. In Latin America, the Association of Water
and Sanitation Regulatory Entities of the Americas (ADERASA) is conducting

190
an international performance comparison to provide better information to regula-
tors and policy-makers. In South East Asia, the Southeast Asian Water Utilities

Network (SEAWUN) is benchmarking its member water suppliers. In Africa, the

Water Utility Partnership (WUP) performs a similar role. In Australia, a National
Benchmarking Framework for Urban Water Utilities (Government of Australia,
2006) has been drawn up under the auspices of the Governments National Water

Commission.

12.3 Data Description

The IBNET database consists of water and wastewater indicators, gathered

from utilities throughout many countries. The data are yearly figures covering
the period 2000 to 2005. The database is evolving as more countries and utilities
participate. The indicators cover a wide range of performance categories including
service coverage, quality of service, consumption and production, non-revenue
water, cost and staffing, billing and collections, financial performance, metering,

assets, network performance, and consumer affordability. The sample selected for
this analysis has been drawn from Asian and Australian suppliers. To minimise
the need for data imputation, only those utilities with substantially complete data
records have been selected. Where imputation is required this has been achieved

by use of additional source material such as annual reports and the ADBs Water
Utility Data Book (McIntosh and Yniquez, 1997). Additionally, an average of the
annual indicator values was taken, because few records contained a complete six
years of annual data.

Table 12.1 shows the indicators and summary statistics for the benchmarking
sample. Table 12.2 shows the sample means disaggregated by country. Country
codes are listed in Table 12.3. Although our objective is not to compare countries,
this table is quite revealing in itself. The sample means of operating cost, wage

191
cost, and per capita production for Australia are the highest in the sample. For
non-revenue water Australia has the lowest sample mean. As production less

non-revenue water approximately equals consumption, this implies that the mean
per capita consumption in Australia is the highest of all countries in the sample.
Note that production/consumption in the IBNET database include residential,
commercial and institutional users. Cambodia and Malaysia also have high levels

of per capita output but with higher rates of non-revenue water than shown by
the Australian statistic.

192
 TABLE 12.1

IBNET Performance Indicators: Summary Statistics 2000-2005

Variable Description Mean Median Min. Max. SD.

COST operating cost per m3 produced 0.212 0.150 0.050 0.800 0.155
W wage cost per m3 produced 0.057 0.045 0.009 0.251 0.042
193

X litres per capita per day output 293.85 207.46 80.56 1129.89 208.28
QU AL hours of service per day 19.7 23.8 1.5 24 6.8
M ET ER meter coverage (% of connections) 90.7 100 0 100 23.6
N RW non-revenue water (% of production) 30.34 32.08 1.00 69.33 12.67
COST RAT revenue cost coverage ratio 1.28 1.30 0.18 3.43 0.57
DU M CLO chlorinated supply (Y=1,N=0) 0.24 0 0 1 0.43

all costs are in USD, inflation adjusted.

 TABLE 12.2

IBNET Performance Indicators: Sample Means Comparison 2000-2005

Indicator Sample AUS CHN IDN IND KHM LAO MYS PHL THA VNM

utilities 114 13 10 15 17 1 1 15 1 1 40
COST operating cost 0.21 0.52 0.15 0.17 0.22 0.11 0.05 0.26 0.24 0.12 0.13
wage cost 0.06 0.10 0.03 0.05 0.08 0.02 0.01 0.05 0.08 0.04 0.05
194

W
X output (lpc) 294 598 182 207 190 592 81 467 209 210 237
QU AL hours service 19.7 24.0 22.6 22.4 5.3 24.0 24.0 24.0 23.3 24.0 20.8
M ET ER meter coverage 91 100 99 100 52 100 100 100 100 100 94
N RW non-revenue 30.3 11.8 20.2 32.3 32.0 21.5 28.0 35.4 29.0 32.8 35.8
COST RAT coverage ratio 1.3 1.8 1.2 1.2 0.6 2.7 1.2 1.1 1.6 1.5 1.4
DU M CLO chlorine 0.2 0 0.5 0.8 0 1 1 0.1 1 1 0.1

costs are in USD, inflation adjusted.

 TABLE 12.3

IBNET Country Codes

AUS Australia LAO Laos

CHN China MYS Malaysia
IDN Indonesia PHL Philippines
IND India THA Thailand
KHM Cambodia VNM Vietnam

12.4 Estimation Results

We have estimated the parameters of two log-linear cost models. The first
bears some similarity to the model used by Estache and Rossi (2002) for the
purposes of performance comparison. Due to data differences however, the same
variables and units cannot be used. Regressors were expressed in log form based

on the observed sample distribution. Model SFA1 is:

ln COST = 0 + 1 ln W + 2 ln X + 3 MET ER + 4 QUAL + 5 DUMCLO

The second model omits two variables that were found not to be significant
in the first and adds two additional variables that were available in the data and

considered to be relevant to efficiency. Model SFA2 is:

ln COST = 0 + 1 ln W + 2 ln X + 3 MET ER + 4 NRW + 5 CRAT IO

Table 12.4 presents the Stochastic Frontier Analysis estimates of models SFA1
and SFA2 using OLS and MLE methods. Standard errors are shown to three

195
significant digits, estimates shown to the same number of decimal places. Signifi-
cant variables are flagged by appending the standard error with an asterisk. The

software programme used was Frontier V4.1 (Coelli, 1997).

12.5 Discussion of Results

In Model SFA1, the QUAL variable is not a significant determinant of COST ,

nor is the DUMCLO dummy variable. In Model SFA2 we omitted QUAL and
DUMCLO and added NRW and CRAT IO. All the variables in this model

are significant. An additional model, not shown here, that included the omitted
variables QUAL and DUMCLO was also estimated with no change in their sig-
nificance. The signs and magnitudes of the estimates of labour and output are
as expected for both models. Metering, which was not in log form, indicates that

a one percent increase in the rate of metering results in about 0.3 cents per m3
increase in operational cost. The sign of NRW in Model SFA2 indicates that
COST decreases with increasing NRW as proportion of total production. This is
a result of simultaneity between the two, as the more appropriate way to express
this is that NRW rates decrease with increased expenditure on leak detection and

maintenance. The operating coverage ratio of revenue to cost would be expected

to fall as the denominator (COST ) increases.
The efficiency of each firm is the ratio of expected cost with the firm efficiency
effect to cost without the effect (Battese and Coelli, 1988):

E (COSTi|ui, xi )
T Ei =
E (COSTi|ui = 0, xi)
= exp(ui )

196
 TABLE 12.4

Stochastic Frontier Analysis: Cost Function

Model: SFA1 OLS MLE

Parameter Estimate Std.Error Estimate Std.Error

0 -2.500 0.457* -2.089 0.522*

W 0.5975 0.0656* 0.6638 0.0714*
X 0.4322 0.0726* 0.3257 0.0819*
MET ER 0.00267 0.00186* 0.00183 0.00167*
QUAL -0.00271 0.00714 -0.00020 0.00690
DUMCLO 0.0169 0.0901 0.0323 0.0788

2 0.1369 0.2943 0.0655

0.8677 0.0871
loglik -45.320 -42.874

Model: SFA2 OLS MLE

Parameter Estimate Std.Error Estimate Std.Error

0 -2.263 0.405* -2.186 0.434*

W 0.5888 0.0579* 0.6266 0.0607*
X 0.4691 0.0623* 0.3881 0.0730*
MET ER 0.00320 0.00149* 0.00387 0.00134*
NRW -0.01048 0.00262* -0.00787 0.00268*
CRAT IO -0.19246 0.0639* -0.2143 0.0602*

2 0.1153 0.2514 0.0573

0.8741 0.0884
loglik -35.55 -33.17

NB: is the ratio-variance parameter u2 / (v2 + u2 )

197
 Table 12.5 shows the estimated efficiency scores for each utility based on each
estimated model. For a cost model with dependent variable in logs, the scores

range from 1 at the frontier, to infinity. In view of the sensitivity to model

specification and formulation of the efficiency statistic, a clustering approach was
considered appropriate and the results have been grouped into 10% deciles. The
highest deciles have the highest efficiency scores and can be regarded as benchmark

leaders in cost efficiency subject to the models used. These included, in Australia,
the Central Gippsland Region Water Authority; in China the Yuncheng Water
Supply Company LT; in Indonesia PDAM Kota Malang, PDAM Kota Padang
Panjang, PDAM Tirta Marta Yogyakarta, PDAM Tirta Sukapura Kabupaten

Tasikmalaya, and PDAM Tirtanadi. In India PHED - Jaipur, PWSSB - Bathinda,

and PWSSB - Dasuya; while in Malaysia, Pulau Pinang, and Terengganu were in
the top decile. In Vietnam, Da Nang Water Supply Co., Ha Giang Water Supply
and Sewerage Co., Khanh Hoa Water Supply and Sewerage Co., Lam Dong Water
Supply Co., Long Khanh Water Enterprise, Quang Ngai Water Supply Co., Quang

Tri Water Supply and Sewerage Co., and the Water Supply Co. of Ba Ria - Vung
Tau Province were in the top decile.
We can observe that the Australian utilities City West, South East Water,
Hunter Valley and Yarra Valley rated in the lowest decile of efficiency scores in

both models. The Chinese utilities Shenzhen and Tianjin also were placed in the
lowest decile for both models. Other consistently low ranked utilities included
PDAM Bandarmasih Banjarmasin in Indonesia, BWSSB in India, and Johor in
Malaysia. Some utilities responded negatively to the changed model (inclusion of

NRW or coverage ratio lowered the efficiency score). These included SA Water
in Australia and CMWSSB in Indonesia. For others there was an improvement,
Labuan and Sibu Water Board in Malaysia for example.
In general we can observe that this form of benchmarking analysis is a powerful

198
tool that is also sensitive to specification and data. In practice any organisation
that engages in benchmarking will need to back up its results and be able to

show why a utility achieved some score. There may be unaccounted for effects
that are specific to some countries, making it difficult to carry out comparative
benchmarking such as this example. Comparisons at the domestic or even regional
level are less likely to have these differences, making performance based pricing

more acceptable. If the utilities trust the method and reliability of the estimates,
and the process is transparent, it is reasonable to expect that they will respond
with increased efforts to improve efficiency.
Price determination based on performance means that the most efficient firms

are used as model firms for the purpose of setting tariffs. The procedure depends
on the regulatory environment. For rate of return regulation, the firms operating
at the frontier would form the basis for determination of allowable costs and the
agreed rate of return on capital. For price caps, the adjustment factor X could
be based on firm level efficiency or the mean firm efficiency. Marginal cost pricing

would entail determination of the marginal cost function of the most efficient firms
and using that to set price. Less efficient firms (with higher marginal cost) would
have an incentive to reduce costs, and possibly change their production technology
to ensure their financial viability. The regulator may also be able to subsidise the

less efficient firms with price increases for a period of time subject to measurable
increases in efficiency.
In practice when performance measurement is extended to price setting, the
group of model firms may need to be restricted both to a geographical region

and utility size. This is to avoid heterogeneity in utility size, regulatory environ-
ment, and service catchment area that are not explicitly accounted for in the cost
function. The Victorian rural and regional water businesses are a good example
of a reasonably homogeneous group of which one or two utilities might become

199
benchmark price leaders. These businesses marginal supply cost might become
the basis of a uniform pricing scheme applicable to all firms, with individual firms

compensated for service area, population growth, and other factors unrelated to
efficiency. This is not dissimilar to the X-factor in CPI-X pricing and presents the
same problem of setting the compensation value. Ultimately it may be the case
that once the benchmark price is set, each utility will be in a position to estimate

its revenue shortfall and appropriate measures can be taken to either reduce costs
or recover the losses from consumers or the owners. More study is required before
a specific model can be selected, however, besides the expected efficiency gains,
this approach does address the criticism that disparate tariff schemes in the same

geographical region are undesirable for consumers.

12.6 Conclusion

In this case study we have presented an example of the use of Stochastic

Frontier Analysis for benchmarking cost efficiency using a sample of water utilities
located in Australia and Asia. The main objective has been to demonstrate the
viability of this approach for urban water pricing. The Stochastic Frontier Analysis
allows us to drop the assumption that all firms are cost minimisers and instead

assumes that some firms are more cost efficient than others. The parameters
of the model when using the maximum likelihood method include a measure of
the deviation from the cost frontier and therefore a normalised measure of cost
efficiency can be estimated. This can be used to rank or cluster firms by efficiency,
or as we have proposed here, to identify the firms whose cost structures would be

used to set prices under either a rate of return form of regulation, price caps or
marginal cost.

200
 TABLE 12.5

Stochastic Frontier Analysis: Cost Efficiency Scores

Efficiency Decile
Country Utility SFA1 SFA2 SFA1 SFA2

AUS Barwon Region Water Authority 1.47 1.31 5 6

AUS Central Gippsland Region Water Au- 1.29 1.14 7 10
thority
AUS Central Highlands Region Water Au- 1.42 1.30 5 7
thority
AUS City West Water Limited 3.11 2.60 1 1
AUS Coliban Region Water Authority 2.28 2.02 2 2
AUS Goulburn Valley Water 2.06 1.69 2 3
AUS Hunter Water Corporation 2.93 2.50 1 1
AUS Power and Water Corporation 2.02 1.78 2 3
AUS South Australian Water Corporation 2.26 2.50 2 1
AUS South East Water Limited 2.72 2.36 1 1
AUS Sydney Water Corporation 1.95 1.75 2 3
AUS WA Water Corporation 1.38 1.60 6 3
AUS Yarra Valley Water Limited 3.21 2.78 1 1
CHN Ankang Water Company 1.89 1.45 2 5
CHN Bozhou Water Company 1.70 1.61 3 3
CHN Haerbin Water Company 1.91 1.80 2 2
CHN Handan Water Company 1.60 1.44 4 5
CHN Hangu Water Company 1.40 1.38 6 5
CHN Jiaozuo Water Company 1.30 1.24 7 8
CHN Shenzhen Water Company 2.90 2.51 1 1
CHN Tianjin Water Company 3.41 3.22 1 1
CHN Weifang Water Company 1.97 1.73 2 3
CHN Yuncheng Water Supply Company 1.21 1.11 9 10
LTD.
IDN PDAM Bandarmasih Banjarmasin 2.32 2.45 1 1
IDN PDAM Kabupaten Banyumas 2.10 1.99 2 2
IDN PDAM Kabupaten Pandeglang 1.20 1.19 9 9

201
 TABLE 12.5

Continued

Efficiency Decile

Country Utility SFA1 SFA2 SFA1 SFA2

IDN PDAM Kabupaten Purwakarta 1.27 1.22 7 8
IDN PDAM Kabupaten Sleman 1.50 1.43 5 5
IDN PDAM Kota Makassar 1.71 1.80 3 2
IDN PDAM Kota Malang 1.19 1.17 10 9
IDN PDAM Kota Padang Panjang 1.27 1.12 8 10
IDN PDAM Kota Pangkal Pinang 1.61 1.56 4 4
IDN PDAM Kota Surakarta 1.42 1.26 5 7
IDN PDAM Tirta Marta Yogyakarta 1.14 1.21 10 9
IDN PDAM Tirta Pakuan Kota Bogor 1.20 1.19 9 9
IDN PDAM Tirta Sakti Kab. Kerinci 1.82 1.69 3 3
IDN PDAM Tirta Sukapura Kabupaten 1.13 1.14 10 9
Tasikmalaya
IDN PDAM Tirtanadi 1.13 1.14 10 10
IND AMC 1.82 1.96 3 2
IND BMC - HED 1.50 1.60 4 3
IND BWSSB 2.32 2.40 1 1
IND CMWSSB 1.85 2.48 3 1
IND DJB 1.32 1.33 7 6
IND MCC 1.51 1.27 4 7
IND PHED - Bharatpur 1.58 1.41 4 5
IND PHED - Bikaner 1.45 1.28 5 7
IND PHED - Jaipur 1.22 1.13 9 10
IND PHED - Jodhpur 2.12 1.88 2 2
IND PHED - Udaipur 1.88 1.40 3 5
IND PWD - Goa 1.34 1.30 6 7
IND PWSSB - Bathinda 1.25 1.19 8 9
IND PWSSB - Dasuya 1.25 1.14 8 10
IND PWSSB - Dera Bassi 1.29 1.17 7 9
IND PWSSB - Gurdaspur 1.33 1.50 6 4

202
 TABLE 12.5

Continued

Efficiency Decile

Country Utility SFA1 SFA2 SFA1 SFA2

IND PWSSB - Patran 1.21 1.24 9 8
KHM Phnom Penh Water Supply Authority 1.27 1.36 7 6
LAO Vientiane City Water Supply Enterprise 1.55 1.48 4 4
MYS Johor 2.34 2.40 1 1
MYS Kelantan state 1.22 1.22 9 8
MYS Kuching 1.45 1.34 5 6
MYS Labuan 2.92 2.23 1 2
MYS LAKU 1.80 1.62 3 3
MYS Melaka 1.66 1.54 3 4
MYS Negeri Sembilan 1.47 1.52 5 4
MYS Pahang State 1.35 1.26 6 7
MYS Perak State 1.69 1.53 3 4
MYS Perlis 1.58 1.45 4 5
MYS Pulau Pinang 1.19 1.14 10 10
MYS Sabah 2.22 2.19 2 2
MYS Sarawak 1.57 1.25 4 7
MYS Sibu Water Board 2.51 2.36 1 2
MYS Terengganu 1.14 1.14 10 10
PHL Silay City Water District 1.40 1.46 5 4
THA Universal Utilitites 1.24 1.25 8 7
VNM Bac Kan Water Supply and Sewerage 1.28 1.49 7 4
Co.
VNM Bac Ninh Water Supply and Sewerage 1.31 1.42 7 5
Co.
VNM Ben Tre Water Supply and Sewerage 1.25 1.25 8 7
Co.
VNM Binh Duong Water Supply and Sewer- 1.36 1.37 6 6
age Co.
VNM Binh Thuan Water Supply and Sewer- 1.26 1.23 8 8
age Co.

203
 TABLE 12.5

Continued

Efficiency Decile

Country Utility SFA1 SFA2 SFA1 SFA2

VNM Cao Bang Water Supply Co. 1.25 1.21 8 9
VNM Da Nang Water Supply Co. 1.14 1.13 10 10
VNM Dong Nai Water Supply and Construc- 1.23 1.25 9 8
tion Co.
VNM Dong Thap Urban Water Supply, Sew- 1.45 1.38 5 5
erage and Environment Co.
VNM Ha Dong Water Supply Co. 1.81 1.86 3 2
VNM Ha Giang Water Supply and Sewerage 1.13 1.14 10 10
Co.
VNM Ha Noi Clean Water Business Co. 1.26 1.23 8 8
VNM Ha Tinh Water Supply Co. 1.19 1.18 10 9
VNM Hai Duong Water Supply Co. 1.64 1.90 3 2
VNM Ho Chi Minh Water Supply Co. 1.28 1.23 7 8
VNM Hoa Binh Water Supply and Sewerage 1.20 1.23 9 8
Co.
VNM Khanh Hoa Water Supply and Sewerage 1.10 1.13 10 10
Co.
VNM Kien Giang Water Supply and Sewerage 1.34 1.39 6 5
Co.
VNM Lam Dong Water Supply Co. 1.14 1.14 10 10
VNM Lang Son Water Supply Co. 1.35 1.33 6 6
VNM Long Khanh Water Enterprise 1.16 1.22 10 8
VNM Nghe An Water Supply Co. 1.43 1.38 5 5
VNM Ninh Thuan Water Supply Co. 1.62 1.75 4 3
VNM Phu Tho Water Supply Co. 1.24 1.19 8 9
VNM Quang Binh Water Supply and Sewer- 1.24 1.34 8 6
age Co.
VNM Quang Ngai Water Supply Co. 1.20 1.13 9 10
VNM Quang Tri Water Supply and Sewerage 1.08 1.06 10 10
Co.

204
 TABLE 12.5

Continued

Efficiency Decile

Country Utility SFA1 SFA2 SFA1 SFA2

VNM Soc Trang Water Supply Co. 1.35 1.24 6 8
VNM Son Tay Water Supply Co. 1.30 1.38 7 6
VNM Tam Ky Quang Nam Water SupplyCo. 1.28 1.34 7 6
VNM Thai Binh Water Supply Co. 1.35 1.27 6 7
VNM Thai Nguyen Water Supply Co. 1.41 1.49 5 4
VNM The Water Supply and Sewerage Co. of 1.39 1.54 6 4
Binh Phuoc Province
VNM The Water Supply Co. of Ba Ria - Vung 1.06 1.08 10 10
Tau Province
VNM Thua Thien Hue Water Supply and 1.26 1.26 8 7
Sewerage Co.
VNM Thua Thien Hue Water Supply and 1.19 1.34 10 6
Sewerage Co.
VNM Tra Vinh Water Supply and Sewerage 1.50 1.55 4 4
Co.
VNM Tuyen Quang Water Supply and Sewer- 1.23 1.30 9 6
age Co.
VNM Vinh Phuc No 2 - Water Supply - Sew- 1.59 1.60 4 3
erage and Environment Co.
VNM Water Supply Enterprise of Long Xuyen 1.20 1.19 9 9
City

205
 CHAPTER 13

Conclusion and Directions for Future Research

13.1 Resume of this Research

We began this thesis by asking some questions about marginal cost water

pricing in an urban context. How is the marginal cost of water determined?

How is the price of water established using marginal cost as the basis? How do
current water prices compare with estimated price based on marginal cost? What
is the welfare impact of current water pricing relative to a proposed marginal
cost based price? In attempting to answer these questions we have examined

both theoretical and practical issues. We began with a selective review of the
theoretical and applied basis for water pricing, examining different aspects of this
theory including cost, demand and welfare.
With cost we have relaxed the assumption of constant long run returns to

scale of production and derived variable returns to scale minimum cost functions
that can be used in empirical work. The lumpy nature of investment in water
infrastructure means that utilities are at any one time likely to produce under
increasing or decreasing returns rather than constant returns, and that marginal

cost is not constant, even over a considerable length of time.

We have considered the demand for water in the estimation of marginal cost.
General practice is to determine long run demand based on historical records of
average and seasonal consumption. This information, coupled with population

206
and demographic data is used for investment planning. Pricing water on the ba-
sis of knowledge of a demand function is not common practice, although block

rate pricing - charging higher marginal rates for water as consumption increases
- is becoming more widespread. In examining demand we focused on modelling
a consumer demand function subject to a block rate tariff. In this case the bud-
get constraint is piecewise linear, and normal assumptions about linearity of the

demand curve no longer hold. We derived the likelihood function for this model
based on use of a two error specification for the demand function.
We presented a method for determination of marginal cost based on mean fac-
tor prices and output, and on a partial equilibrium of marginal cost and the de-

mand curve. We examined the welfare loss arising from pricing at below marginal
cost and the level of subsidisation required at marginal cost when there are in-
creasing returns to scale.
Two case studies applied this theory to data sets drawn from Australia and
the Philippines. The first looked at cost and welfare in the supply of water to

domestic users in Victoria. Our conclusion was that urban water is produced
under increasing returns to scale - commonly this is associated with regulated
industries and the need to maintain essential supply reserves. We determined
that the current volumetric rate for metropolitan water businesses was around

26% less than the actual marginal cost of water, while for regional businesses it
was 35% less. If marginal cost pricing were to be used, a subsidy of about of 3.3%
of current revenues would still be required for metropolitan businesses, while for
regional businesses a subsidy of 7.2% would be required.

The second case study was focused on recent events in water privatisation in
Manila. The government owned water utility was considered to be inefficient and
the infrastructure in need of significant investment to cater to the demands of
urban population growth. Privatisation was expected to reduce the burden of

207
debt that had been accumulated by the government in operating the water and
sewerage services. Two firms were awarded contracts for the supply of domestic

and commercial water in 1997, and the department previously responsible for
water supply assumed the role of regulator. Estimates of cost functions indicated
decreasing returns to scale, although the hypothesis of constant returns was not
rejected. Household demand was estimated using two methods. First we used

a sample of households consisting of those with a water connection and those

that relied on other sources such as water carters. A weighted average price
was determined and least squares used to estimate the parameters of a log-linear
demand model. The second method involved maximum likelihood estimation of

household demand using a two error model derived in Chapter 4. The sample
consisted only of households with a water connection. The results of these revealed
that, for the mixed sample, price was less elastic than for the sample of connected
households. We interpreted this to mean that those households with a water
connection pay an overall lower price than the average household, and consume

more. They are able to reduce consumption when faced with higher prices because
essential requirements can still be met. The welfare effects of a price increase for
households with a connection were found to be positive, and in 2000 average
prices were about 26% below marginal cost at the efficient output level. Prices

have increased significantly since 2001, however the basis for price determination
by the regulator continues to be governed directly by the concessionaires costs
without consideration of impacts on consumer welfare or predictions of changes
in demand.

Estimation results have been discussed independently in each case study and
jointly in Chapter 11. There we have contrasted both sets of results with regards to
scale economies, elasticities, pricing, tariffs, ownership, and regulation. Following
on, we presented a contrasting approach to marginal cost and pricing that relaxes

208
assumptions about cost minimisation, assumptions that form the basis for the
cost models used in the case studies. We presented a model that allows us to

estimate the deviation of firms from a performance frontier, in contrast to an

estimate of their mean cost. This was followed by an applied example that used
benchmarking data from the World Bank to estimate a cost frontier for a sample
of water utilities throughout Australia and Asia. By assignment of efficiency

scores to clusters (deciles) we proposed that the higher performing utilities, as

well as providing leadership in efficiency, would form the basis of the information
used by regulators to price the service. Performance based pricing such as this
would encourage firms to become more efficient as their prices would effectively

be capped at the marginal cost of the most efficient firms.

13.2 Originality of this Contribution

In the introduction to this thesis we previewed the novelty of our approach

in three areas: the use of functional forms to describe the key variables: cost,

demand, and welfare; the recognition that water utilities often operate with in-
creasing returns to scale and the implications of this in respect of cost recovery
and over use of the resource; and the acknowledgement that utilities may not
behave as cost minimisers in all cases, particularly if operating under rate of re-

turn regulation. We have considered solutions to these problems in the form of

Ramsey pricing and an alternative pricing mechanism based on use of efficiency
frontiers. These areas are commonly dealt with independently but seldom in a
unified manner. Additionally, we have taken consumer demand into account; as

part of the estimation of efficient levels of output, and in examining the properties
of consumer demand subject to a increasing block rate tariff.
This work has emphasised an alternative approach to conventional water price
and demand analysis based on the use of econometric modelling. The econometric

209
models that we have presented in this thesis are characterised by their parsimony
in respect of choice of regressors, an approach that follows after Varian (1984,

Ch. 4). The advantage of this approach is in the relative ease of interpretation of
results in comparison to the case when there are many regressors, some of which
are likely to be correlated. The use of econometric models, in contrast to estab-
lished methods involving discounted cost and demand, has produced useful results

that will improve with better data and more research in the models used. Most
importantly this approach allows us to measure parameters of water production
and consumption including economies of scale, cost and output elasticities, and
price and income elasticities of demand. By use of sensitivity analysis we have

shown that the dependent variables (marginal cost and welfare) are not sensi-
tive to the level of output but are sensitive to changes in the coefficients of the
underlying cost and demand functions.
The application of econometric models in the form of three case studies, each
involving a different data set and regulatory environment, forms the main original

applied contribution. These results are expected to be of interest to those engaged

in applied research as well as the involved utilities and regulators.

13.3 Urban Water Pricing Policy

In the introductory chapter when stating the objectives of this research we

also asked: what are the main policy issues that we need to address in a debate
on water pricing - particularly one based on marginal cost? In this section we will
attempt to provide some answers to this question based on the experiences and

findings of this research. The detail of empirical findings upon which some of these
remarks are based can be found at the end of each case study and at the beginning
of Chapter 11. A broader water policy discussion is beyond the scope of this work
as it would need to address all of the uses of water including agricultural, mining,

210
cultural, recreational, and ecological. The interested reader is referred to reports
that set out policy directions in the global context with emphasis on developing

parts of the world: United Nations (2006); and specifically in the Australian
context: Business Council of Australia (2006); Marsden and Pickering (2006);
Young, Proctor, Qureshi, and Wittwer (2006).

13.3.1 Tariff Design and Variable Unit Pricing

For regulators interested in setting price, or price caps, or influencing tariff

design, an understanding of the scale economies and the marginal cost function of
the utilities is essential. This is because the relationship between marginal cost,
average cost and price will determine: first, whether the utility has to subsidise its
operation by offering its services to different classes of user at different prices or

by direct assistance from its owners; and second, whether the utility is engaging
in monopoly pricing. Because costs change over time, continuous monitoring of
utility production and investment costs, is necessary for the regulator to carry out
this function.
The choice of price and tariff design should be governed by efficiency and

equity concerns and the need for sustainable resource management. To reflect
the presence of scarcity, the marginal cost curve will be increasing and there will
be decreasing returns to scale. If scarcity is not taken into account, and there
are increasing returns to scale, marginal cost pricing, although more efficient in

the economic sense, will result in both over use of water and in revenues below
the total cost of supply. A regulator therefore must price the unknown scarcity
component of marginal cost by adding its user cost so that the marginal cost curve
is increasing. This is achieved with the use of a rising block tariff. The unresolved

difficulty, however, is that the shadow price of scarcity remains difficult to quantify,
in particular in the urban water context.

211
 A rising block tariff will therefore increase revenues for the utility operating un-
der increasing or constant returns to scale relative to production and distribution

costs. These revenues can be used to offset the scarcity premium by investment
in projects that increase future supply, or to subsidise low volume consumers, or
a combination of both. The risk of this approach is that it still does not cap
consumption for users for whom price is unimportant. Therefore, in times when

supplies are under stress, a combined approach that includes restrictions and fi-
nancial penalties for excessive consumption will still be required in addition to
the rising tariff. Where water supplies are not under stress, there is less of a need
to incorporate the scarcity component in marginal cost, but Ramsey pricing will

still be required to avoid financial losses.

Despite their increasing use, rising block rate tariffs are an imperfect instru-
ment for dealing with increasing marginal costs and scarcity. First, determination
of the parameters of a block rate tariff - the block lengths and prices - is difficult
to implement, despite the research that has been devoted to this area. Leaving

this to individual utilities (as is being done in Victoria) places demands on the
available skills and results in a variety of different tariffs for utilities whose pro-
duction and costs are unlikely to be greatly different. Second, a simple one or
two block rate tariff aims to combine market signals (increasing marginal price)

and essential service security (constant unit price for a quota) in one. These two
ought be regarded independently so that the nonessential component becomes
fully marginal cost based. This could be effected in different ways. The nonessen-
tial usage block prices can be reset at regular intervals, to account for changing

seasonal demands and supply constraints. Alternately, the volumetric price could
vary continuously with consumption1 . The application of market pricing in urban
1
The price would be p = + kQ , where Q is consumption in excess of the essential compo-
nent, and , k and are adjustable parameters.

212
water is lagging behind other areas including agricultural water and domestic and
industrial electricity supply. Smart metering in electricity supply where consumers

can modify their consumption and the price they pay according daily variation in
supply and demand is one area that utilities and regulators could investigate to
see how market pricing might work.
The use of an increasing block rate tariff when marginal production and dis-

tribution costs are decreasing affords the opportunity of cross-subsidisation of the

low-cost essential component of water supply. This creates a problem of equity
because this component is offered to all consumers at the same price, regardless
of income or total consumption. High volume consumers would in effect be sub-

sidising part of their own consumption. One possible solution to this would be to
offer two different tariff structures, one with and one without the low-cost essen-
tial component. Average consumption over the last period (not proxied income)
would determine which one was applied, but care would be needed to avoid pe-
nalising large households with normal per capita levels of consumption. We must

emphasise again that this approach would only apply when total marginal costs
are increasing because of scarcity or other supply constraints. Without these con-
straints, there does not appear to be any reason for offering water at anything
other than its (decreasing) marginal supply cost, providing supplies are abundant

and all investment and user costs accounted for.

The point is sometimes made that water prices need to be predictable. This
is interpreted to mean they should be constant and that any variability will have
adverse effects. The use of marginal cost pricing has been criticised for these rea-

sons, as has been the use of seasonal adjustments to price. As efficiency becomes
more important, these arguments become harder to justify and more variability in
price may be necessary. This may be applied only to unit prices for consumption
above the level of the essential component, the price of which could remain fixed

213
for much longer periods. Publication or issuing notices of price adjustments in
advance of their implementation would be one way of ensuring predictability.

13.3.2 Efficiency and Competition

In the later chapters of this thesis we turned our attention to the area of
efficiency measurement in pricing water. For regulators and those who determine
national water policy, efficiency in production and efficiency in consumption are
two areas for consideration. Pricing water according to the production costs of the

most efficient producers reduces the risk of having a price that reflects side-effects
arising from regulatory weakness that does not give the utility an incentive to
reduce costs, or that encourages abuse of monopoly powers. It also makes up for
the absence of a competitive market among peers who will be forced to reduce their

costs. Promoting efficiency in consumption, by smart metering, public awareness

campaigns, or by the use of water saving appliances should also be encouraged.
The other area that competition can play a role is in service provision. This
involves splitting off some non-core services to outside firms who would compete
to offer the lowest cost service. These services include maintenance, meter instal-

lation and reading, accounting and billing systems, water treatment and storage,
waste disposal, management, and even distribution. This is becoming more com-
mon, however competitive bidding does not ensure that the most efficient firm
will win because of the potential for underbidding; although there are solutions

for this, for example as described in Solon and Pamintuan (2000). Generally, more
competitive supply in the urban water supply sector should be encouraged in the
light of evidence that the presence of competition contributes to efficiency.

214
13.3.3 Agricultural Water Use and Trade

It is difficult, particularly in the Australian case, to discuss efficiency in the

urban water sector in isolation. This is because, on a national scale, the urban
sector accounts for only about 12% of consumption while the agricultural produc-

tion sector accounts for almost 70% of consumption (Kaspura, 2006). The greatest
gains nationally, will therefore be made by increasing efficiency in agricultural wa-
ter use. Existing water trading schemes mean that rural water is already priced
at its marginal benefit and therefore there is a strong incentive for users to in-

troduce technological improvements so that efficiency in use is increased. Supply

constraints where they exist will not be resolved through the water market as the
tradeable price does not include the investment cost of increased supply. There is
therefore a need for increased investment in securing agricultural water supply in

the current environment. One idea that has gained some attention in recent times
is the purchase of rural water entitlements for residential use to overcome shortages
in urban areas (Quiggin, 2006). The adoption of this appears to be increasingly
less likely as existing water allocations are reduced because of drought.

13.3.4 Data Standardisation

The use of innovative analytical and statistical approaches, and performance

benchmarking and monitoring is predicated on the availability of quality data.

Many studies have noted that data availability is a considerable constraint in
applied research. A worrying concern is that useful data is kept out of the public
domain; it is considered proprietary or of no research value. Regulators have a

role to play in ensuring that data is available, shared, and conforms to common
standards. In the words of the United Nations (2006):

It is only when the data has been collected and analysed that we can
properly understand the many systems that affect water (hydrological,

215
 socio-economic, financial, institutional and political alike), which have
to be factored into water governance.

Urban water policy needs to address these weaknesses by making water ac-
counting mandatory practice for regulators and utilities, based on standards com-
mensurate with professional practice.

13.3.5 Other Implications of Pricing Policy

Throughout the course of this study we have made the point that welfare losses

are associated with existing water pricing systems in many parts of the world. We
estimated these losses in the Case Studies for Victoria and Manila. This means
that many utilities operate at an economic loss and require subsidisation from their
owners. In the broader sense this inefficiency leads to a undervaluing and over-

use of the resource with the resultant loss to other competing uses. Continuation
of this situation poses a very serious threat to our society when associated with
decreased inflows and population growth.
This and other studies have demonstrated the welfare gains that can be made

by increasing urban water prices and adopting a marginal cost approach to price
determination. This policy is now being put into practice in many parts of the
world including Australia. Although deemed welfare improving, we need to ask
if higher prices create new or additional problems of equity. In Australia these

issues are likely to be addressed through the existing concession schemes for pen-
sioners and welfare recipients that are linked to CPI increases. However, with a
new round of above CPI water price increases likely in most states the next few
years, additional concessional allowances may be required for some. Other equity
improving measures such as the use of Ramsey pricing have in some studies shown

to decrease net welfare (Renzetti, 1992). The requirement for some form of price
discrimination among different classes of user, according to elasticity of demand

216
for example, is however unlikely to diminish with a more efficient pricing policy.
Block rate pricing and connection charges based on property valuations (Sibly,

2006) are potential but partial solutions to this problem. It is important to note
that these proposals are in contrast to the current use of water restrictions which
does not manage scarcity by price signaling and forfeits revenue from those users
who are willing to pay for high levels of consumption. Regardless, some form of

cap on high volume users would still be required when capacity constraints are
present.
Another implication of an increase in water prices based on the long run
marginal cost is that there may be periods when utilities make windfall prof-

its that are difficult to justify to consumers. Indeed this is already the case with
the Victorian state government under criticism for not directly reinvesting profits
from the metropolitan water businesses. The lack of investment in water infras-
tructure in Victoria and other parts of the world might have resulted in this way,
however, as shown in the First Case Study, utilities may be reluctant to borrow

to finance capital investment. There may be political reasons for this, but we
also have shown that for certain models, costs are more sensitive to interest rates
than wages. The implication of this is that debt financing of capital projects by
government owned corporations is unpopular and alternative methods of financing

such as equity ownership and Build Own Operate and Transfer (BOOT) schemes
need to be examined further.
Long run marginal cost pricing also becomes problematic when the factors
that lead to its adoption are no longer relevant. The scarcity premium may not

be required because the investment has been carried out and paid for and capacity
constraints are not present. These issues add to the viability of proposals such as
seasonal based pricing and adjusting prices relative to storage supplies (Grafton
and Kompas, 2006). Additionally, a long term downward price adjustment is not

217
unreasonable if seasonal and long term inflows are adequate. Recently all major
capitals of Australia with the exception of Darwin have decided to build desali-

nation plants, while Perth already has one in operation and is embarking on its
second. Ignoring environmental costs, desalination has become a viable alterna-
tive because its cost is comparable to the long run marginal costs of supplying
surface water. If capacity constraints are not present however and surface water

prices can again be set below the cost of desalinated water, then the plant will
only operate in a standby mode. This is unlikely to be acceptable and therefore
the possibility is that water prices in the future might be governed by the marginal
cost of desalinated water, at least in the Australian capitals.

13.4 Future Research Directions

In conducting this research, we have been presented with alternate paths and
opportunities that appear interesting and potentially rewarding. It is impossible
to explore all of these in the space of a thesis. They relate to both application

and theory.
In applied work, there is a need to more fully account for user costs and
externalities by use of shadow pricing methods such as contingent valuation and
coping cost (Devicienti, Klytchnikova, and Paternostro, 2004). Two data sets have

recently become available that can contribute to our understanding in this area.
The first is a sample of water supply firms from Cambodia, and originates from
research conducted by Basani, Isham, and Reilly (2005). The second is based on
Philippines price indices and presents an opportunity to explore linear expenditure

systems as described in Deaton and Muellbauer (1980). In terms of econometric

methods used, the use of non-linear models, multiple equation estimation, and
multiple output models are potentially useful in applied work, especially as data
sets become more complex, and we seek to explain more complex relationships.

218
 In theory development, the relationship between returns to scale, elasticity, and
welfare is an area where more research is needed. Similarly, capital adjustment

needs to be explained within the cost model, which implies that a dynamic model
is required. More work is needed in the area of water trading and rural-urban
transfers; this work may need to consider opportunity costs when water rights are
traded, including impact on land valuations and conversion to dryland agriculture,

and the impact on agricultural prices.

13.5 Closing Comments

Increasingly there is a need to fully account for the economic cost of water

service including opportunity costs, externalities, and investment in the long run.
To justify future price increases, authorities and regulators need to take these costs
into account so that the economic case for price increases can be made. As part
of this undertaking, a re-examination of the reporting and accounting of water
services needs to take place so that economic costs are quantified, made more

transparent, and employed in decision making to the fullest extent possible.

The estimation and use of functional forms for the analysis of urban water
costs and demand is promising and merits place in an array of tools used by
utilities, regulators, and policy makers. The results that we have obtained with

this approach provide some evidence that urban water has been undervalued in
recent years. The value of water should be reflected in its price and the value that
consumers place on its use. We need to fully understand production, investment
and opportunity costs, the cost of scarcity and externalities, and what impact

price has on demand.

Water is essential to our survival and prosperity. In abundance, it is thought
of as a public good. It is increasingly acknowledged however, that water is an
economic good and needs to be priced accordingly. The objective of this thesis

219
has been to demonstrate the use of marginal cost as an economic basis for pricing
water and to give examples of how this might be applied. The recent evidence is

that valuing water on such as basis is becoming something of an imperative for

our society.

220
 APPENDIX A

Properties of the Variable Returns to Scale Cost Function

In the following we will express the variable returns cost function 3.4 in the
equivalent form:

C = Aw r x

1
where = +
.

1. Homogeneous of degree one in factor prices.

C = A(kw) (kr) x

= k (+) C

= kC

2. Homogeneous of degree in output.

C = Aw r (kx)

= k C

221
3. Nondecreasing in factor prices.

C
= C
w w

= C
w( + )

It follows that this last expression is non-negative from the definitions of

and . By symmetry the same holds for all other factors.

4. Nondecreasing in output.
C
= C
x x

It follows that this last expression is non-negative from the definitions of

and .

5. Concave in factor prices.

C
= Aw 1 r x
w
2C
= A( 1)w 2 r x
w 2
C
= ( 1) 2
w

We only need to examine the term 1 in this last expression.

1 = 1
+

=
+
0

222
Note that the minimum cost function is not concave in output, as:

2C C
= ( 1)
x2 x2

and the term ( 1) has ambivalent sign.

223
 APPENDIX B

Transformation of the Two Error Model Likelihood Function

B.1 Introduction

The purpose of this transformation is to create a new pair of standardised

error variables that are uncorrelated. This enables the use of univariate standard
normal and cumulative density functions in the likelihood function. A set of
identities used in this derivation is included at the end of this Appendix.

B.2 Segment Likelihood

The transformation entails decomposition of the bivariate normal distribution

into the product of conditional and univariate density functions (see Johnston and
DiNardo, 1997, p. 15):

f (vi , i ) = f (vi )f (i |vi )

We begin with the conditional density function which is by definition:

" 2 #
i |v

1 1
f (i |vi ) = exp
| 2 2 |v

The conditional mean is:

|v = v + v = 2 v
v v

224
 where is correlation between v and . The conditional variance is:

2
|v = 2 (1 2 )

Therefore:

!2
2
1 1 i v
f (i |vi ) = p exp p
2(1 2 ) 2 (1 2 )
!2
1 1 i / v/v
= p exp p
2(1 2 ) 2 (1 2 )

Now define three new standardised variables:

zi = vi /v

ti = i /
ti zi
ri = p
(1 2 )

Note that we need to verify that r is a standardised variable. By inspection,

its mean is zero. For the variance:

! ! !
t z t z
var(r) = var p + var p + 2 cov p ,p
1 2 1 2 1 2 1 2
1 2 22
= +
1 2 1 2 1 2
= 1

225
 Therefore we have for the conditional density:

1
f (i|vi ) = p f (ri )
(1 2 )

Now we can reformulate the bivariate density function, with reference to the

identity that converts a normal density into standard normal density:

f (vi , i ) = f (vi )f (i |vi )

1
= p f (zi )f (ri )
v (1 2 )

Finally, we can express the likelihood of the first segment in terms of the new

standard variables:

x gi (1) Z 1i =x gi(1)
1
Z
h(vi , i )di = p f (zi )f (ri )di
v (1 2 )
Z t1i
1
= p f (zi )f (ri )dti
v (1 2 )
Z r1i
1
= f (zi )f (ri)dri
v
1
= f (zi )F (r1i )
v

Note that each change of variable of integration is achieved by differentiation

of the standardised variables and that the limits of integration must also change

when this is carried out. The likelihood for the second segment follows by the
symmetry of the density function.

B.3 Kink Likelihood

The kink likelihood only requires standardisation of the errors:

226
 x gi (2) Z 2i =x gi (2)
1 i 1 i
Z
f (i)f (i )di = f( ) f ( )di
x gi (1) 1i =x gi (1)
Z t2i
1
= f (si ) f (ti )dti
t1i
1
= f (si )[F (t2i ) F (t1i )]

where si is the standardised form of i .

B.4 Identities used in Transformations

B.4.1 Properties of the Error Variables:

vi = i + i
var(v) = var() + var()
cov(, ) = 0

cov(v, ) = 2
cov(v, ) = 2
= corr(v, ) = cov(v, )/(v ) = /v
p
/v = 1 2
p
/ = 2 / 1 2

227
B.4.2 Properties of the Standardised Error Variables:

1 XX
cov(z, t) = zi tj
n1 i j
1 1 XX
= vi j
n 1 v i j
1
= cov(v, )
v
=

= corr(z, t)

= corr(v, )

f (v) = f (v z)
 
1 1 2
= exp (v z) /var(v z)
2var(v z) 2
 
1 1 1
= exp (z)2
v 2 2
1
= f (z)
v

The following identity confirms the use of the conditional density transforma-

tion:

1 XX
cov(z, r) = zi rj
n1 i j
1 XX p 1 XX p
= (zi tj / 1 2 ) (zi zj / 1 2 )
n1 i j n1 i j
= 0

228
 BIBLIOGRAPHY

Aigner, D., C. Lovell, and P. Schmidt. Formulation and estimation of stochastic

frontier production functions models. Journal of Econometrics, 6:2137, 1977.

Australian Broadcasting Commission. Murray-Darling inflows at record lows,

2007. URL http://www.abc.net.au/news/newsitems/200702/s1846428.
htm. Tuesday, February 13, 2007. 11:29am (AEDT).

Australian Bureau of Statistics. Service industries producer price indexes 1968-

2007. Technical Report ABS 6427.0, Australian Bureau of Statistics, 2007.

Averch, H. A., and L. L. Johnson. Behaviour of the firm under regulatory con-
straint. American Economic Review, 52:10531069, 1962.

Balistreri, E. J., C. A. McDaniel, and E. V. Wong. An estimation of U.S. industry-

level capital-labor substitution elasticities: support for Cobb-Douglas. The
North American Journal of Economics and Finance, 14(3):343356, 2003.

Baron, D. P., and R. B. Myerson. Regulating a monopolist with unknown costs.

Econometrica, 50:911930, 1982.

Basani, M., J. Isham, and B. Reilly. Water demand and the welfare effects of
connection: some empirical evidence from Cambodia. Department of Economics
discussion paper (series 2004-29), Middlebury College US., 7 2005.

Battese, G. E., and T. J. Coelli. Prediction of firm-level technical efficiencies

with a generalized frontier production function and panel data. Journal of
Econometrics, 38:387399, 1988.

Berndt, E., and D. Wood. Technology, prices, and the derived demand for energy.
Review of Economics and Statistics, 57:376384, 1975.

Bhattacharyya, A., E. Parker, and K. Raffiee. An examination of the effect of

ownership on the relative efficiency of public and private water utilities. Land
Economics, 70(2):197209, 1994.

Billings, R. Alternative demand model estimators for block rate pricing. Water
Resources Bulletin, 23(2):341345, 1987.

229
Blanciforti, L., and R. Green. An almost ideal demand system incorporating
habits: an analysis of expenditures on food and aggregate commodity groups.
The Review of Economics and Statistics, 65:511515, 1983.
Boardman, A. E., D. H. Greenberg, A. R. Vining, and D. L. Weimer. Cost-benefit
Analysis: Concepts and Practice. Prentice Hall, 2006.
Bougheas, S., and T. Worrall. Cost padding in regulated monopolies. Discussion
Papers in Economics, University of Nottingham., 01/12, 2006.
Bregman, A., M. Fuss, and H. Regev. The production and cost structure of Israeli
industry. Evidence from individual firm data. Journal of Econometrics, 65:
4581, 1995.
Bruggink, T. H. Public versus regulated private enterprise in the municipal water
industry: a comparison of operating costs. Journal of Economics and Business,
22(1):111125, 1982.
Burtless, G., and J. Hausman. The effect of taxation on labor supply: Evaluating
the Gary negative income experiment. Journal of Political Economy, 86(6):
11031130, 1978.
Business Council of Australia. Water under pressure. Australias man-made water
scarcity and how to fix it. Technical report, Business Council of Australia, 2006.
Chicoince, D., and G. Ramamurthy. Evidence on the specification of price in the
study of domestic water demand. Land Economics, 62(1):2632, 1986.
Chotrani, R. Lessons from Philippines water privatization. Asian Water, July
1999.
Coase, R. The marginal cost controversy. Economica, 13(8):169189, 1946.
Coelli, T. J. A guide to FRONTIER Version 4.1: A computer program for stochas-
tic frontier production and cost function estimation. Technical report, Centre
for Efficiency and Productivity Analysis, Department of Econometrics, Univer-
sity of New England, Armidale. NSW., 1997.
Dachraoui, K., and T. M. Harchaoui. Water use, shadow prices and the Canadian
business sector productivity performance. Technical report, Statistics Canada,
Micro-economic Analysis Division, 12 2004.
Dalhuisen, J., R. Florax, H. de Groot, and P. Nijkamp. Price and income elas-
ticities of residential water demand: A meta-analysis. Land Economics, 79(2):
292308, 2003.
David, C., A. Inocencio, R. Abracose, R. Clemente, and G. Tabios. Optimal Water
Pricing in Metro Manila. Paper presented at the Second Workshop on Urban
Water Pricing, March 17, 1998., 1998.

230
David, C. C. MWSS privatisation: Implications on the price of water, the poor,
and the environment. Technical report, Philippine Institute for Development
Studies, April 2000.

David, C. C., and A. B. Inocencio. Understanding household demand for water:

The Metro Manila case. Technical report, EEPSEA Economy and Environment
Program for Southeast Asia, 2002.

David, C. C., A. B. Inocencio, R. S. Clemente, R. P. Abracosa, F. M. Largo, G. Q.

Tabios, and E. L. Walag. Urban water pricing: The Metro Manila and Metro
Cebu cases. Technical report, Philippine Institute for Development Studies,
July 2000.

Deaton, A. The Analysis of Household Surveys. Johns Hopkins University Press,

1997.

Deaton, A., and J. Muellbauer. An almost ideal demand system. The American
Economic Review, 70:312326, 1980.

Deaton, A., and J. Muellbauer. Economics and Consumer Behavior. Cambridge

University Press: Cambridge, 1983.

Devicienti, F., I. Klytchnikova, and S. Paternostro. Willingness to pay for water

and energy: an introductory guide to contingent valuation and coping cost
techniques. World Bank Group Energy Working Notes, 3:139, 2004.

Dougherty, C. Introduction to Econometrics. Oxford University Press, 1992.

Esguerra, J. The corporate muddle of Manilas water concessions. Wateraid and

Tearfund, 2003. http://www.wateraid.org.uk.

Estache, A., and M. A. Rossi. How different is the efficiency of public and private
water companies in Asia? The World Bank Economic Review, 16(1):139148,
2002.

Faria, R. C., G. da Silva Souza, and T. B. Moreira. Public versus private water
utilities: Empirical evidence for Brazilian companies. Economics Bulletin of the
Catholic University of Brasilia, 8(2):17, 2005.

Feigenbaum, S., and R. Teeples. Public versus private water delivery: a hedonic
cost approach. Review of Economics and Statistics, pages 672678, 1983.

Fletcher, R., and C. M. Reeves. Function minimization by conjugate gradients.

Computer Journal, 7:148154, 1964.

Forero, J. Who will bring water to the Bolivian poor? The New York Times, 12
2005.

231
Fox, W. F., and R. A. Hofler. Using homothetic composed error frontiers to
measure water utility efficiency. Southern Economic Journal, 53(2):461477,
1986.

Gaudin, S., R. Griffin, and R. Sickles. Demand specification for municipal water
management: Evaluation of the Stone-Geary form. Land Economics, 77(3):
399422, 2001.

Government of Australia. Draft national benchmarking framework: Performance

reporting model for urban water utilities. Technical report, National Water
Commission, 2 2006.

Government of South Australia. Waterproofing Adelaide 2005-2025, 2005.

Grafton, R. Q., and T. Kompas. Sydney Water: Pricing for sustainability. Tech-
nical report, Economics and Environment Network, Australian National Uni-
versity, 2006. URL http://een.anu.edu.au/download_files/een0609.pdf.

Green, R., and J. M. Alston. Elasticities in AIDS models. American Journal of

Agricultural Economics, 72:442445, 1990.

Greene, W. H. Econometric Analysis. Prentice Hall, 2003.

Harris, J., D. Tate, and S. Renzetti. Economic principles and concepts as ap-
plied to municipal water utilities. Technical report, GeoEconomics Associates
Incorporated, 2002.

Hausman, J. Exact consumers surplus and deadweight loss. American Economic

Review, 71(4):662676, 1981.

Henningsen, A., and J. D. Hamann. Systemfit: A package to estimate simulta-

neous equation systems in R. Technical report, University of Kiel and Forest
Informatics, Inc., 3 2006.

Hewitt, J., and W. Hanemann. A discrete/continuous choice approach to residen-

tial water demand under block rate pricing. Land Economics, 71(2):173192,
1995.

Hill, R., W. Griffiths, and G. Judge. Undergraduate Econometrics 2nd Ed. John
Wiley, 2001.

Hines, L. G. The long-run cost function of water production for selected Wisconsin
communities. Land Economics, 45(1):133140, 1969.

Hotelling, H. The general welfare in relation to problems of taxation and of railway

and utility rates. Econometrica, 6(7):242269, 1938.

Inocencio, A. B. Public-private partnerships in Metro Manila, Philippines. In 3rd

World Water Forum, Osaka, 19 March, 2003, PPCPP Session, 2003.

232
Jehle, G. A., and P. J. Reny. Advanced Microeconomic Theory. Addison Wesley,
2001.

Johnston, J., and J. DiNardo. Econometric Methods (International Edition).

McGraw-Hill, 1997.

Jondrow, J., C. K. Lovell, I. S. Materov, and P. Schmidt. On the estimation

of technical inefficiency in the stochastic frontier production function model.
Journal of Econometrics, 19:233238, 1982.

Kaspura, A. Water and Australian cities: Review of urban water reform. Technical
report, National and International Policy Unit, Engineers Australia, 2006.

Keating, M. Water pricing and its availability. Technical report, Independent

Pricing and Regulatory Tribunal, NSW State Government, 2006.

Kim, H. Y. The translog production function and variable returns to scale. The
Review of Economics and Statistics, 74:546552, 1992.

Kim, H. Y. Marginal cost and second-best pricing for water services. Review of
Industrial Organization, 10:323338, 1995.

Klein, M. W. Mathematical Methods for Economics. Addison-Wesley, 1998.

Kolstad, C. D., and J.-K. Lee. The specification of dynamics in cost function and
factor demand estimation. The Review of Economics and Statistics, 75:721726,
1993.

Laffont, J.-J., and J. Tirole. Using cost observation to regulate firms. Journal of
Political Economy, 94(3):614641, 1986.

Laffont, J.-J., and J. Tirole. The politics of government decision making. a theory
of regulatory capture. Quarterly Journal of Economics, 106(4):10891127, 1991.

Layard, P., and A. Walters. Microeconomic Theory. McGraw-Hill, 1987.

Littlechild, S. Economic regulation of privatised utilities and some further reflec-

tions. Oxford Review of Economic Policy, 4(2):4068, 1988.

Luenberger, D. G. Microeconomic Theory. McGraw-Hill, 1995.

Magdahl, J., H. Srreimd, E. Christensen, A. Preston, T. Kronen, and O. Berg.

Privatisation of water. Public-private partnerships: do they deliver to the poor?
Technical report, Norwegian Forum for Environment and Development, 4 2006.

Mann, P. C. Water utility regulation: rates and cost recovery. Reason Foundation,
Policy Study, 155, 1993.

233
Marsden, J., and P. Pickering. Securing Australias urban water supplies: op-
portunities and impediments. Technical report, Marsden Jacob Associates, 11
2006.

McIntosh, A. C., and C. E. Yniquez. Second Water Utilities Data Book - Asian
and Pacific Region. Asian Development Bank, 1997.

Moffitt, R. The econometrics of piecewise-linear budget constraints. Journal of

Business & Economic Statistics, 4(3):317328, 1986.

Moffitt, R. The econometrics of kinked budget constraints. Journal of Economic

Perspectives, 4(2):119139, 1990.

Montemayor, C. A. The Manila water privatization fiasco and the role of Suez
Lyonnaise/Ondeo. Presentation for the Summit for Another World, Evian,
2003, 2003.

National Statistics Office. Technical notes on the Family Income and Expenditure
survey. Technical report, National Statistics Office, Manila, Republic of the
Philippines, 2000a. http://www.census.gov.ph/fies.

National Statistics Office. 2000 Census of Population and Housing. Technical

report, National Statistics Office, Manila, Republic of the Philippines, 2000b.

Nelder, J., and R. Mead. A simplex algorithm for function minimization. Computer
Journal, 7:308313, 1965.

Nelson, J. A. Estimation of food demand elasticities using Hicksian composite

commodity assumptions. Quarterly Journal of Business and Economics, 33(3):
5168, 1994.

Nerlove, M. Returns to scale in electricity supply. In Christ, C., editor, Measure-

ment in Economics: Studies in Mathematical Economics and Econometrics.
Stanford University Press, 1963.

Newbery, D. M. The New Palgrave Dictionary of Economics and the Law, chapter
Rate-of-return regulation versus price regulation for public utilities. MacMillan,
1998.

Nieswiadomy, M., and D. Molina. Comparing residential water demand estimates

under decreasing and increasing block rates using household data. Land Eco-
nomics, 65(3):280289, 1989.

Perez, V. The failed water privatisation in Manila. Technical report, Japan Center
for a Sustainable Environment and Society, 2003.

Pindyck, R. S., and D. L. Rubinfeld. Econometric Models and Economic Forecasts.

McGraw-Hill, inc., 1991.

234
Pollack, R. A., and T. J. Wales. Estimation of complete demand systems from
household budget data: The linear and quadratic expenditure systems. Amer-
ican Economic Review, 68:348359, 1978.

Price, J. Comparing the cost of water delivered. Technical report, Office of Water
Services, UK Government, March 1993.

Pushpangadan, K., and G. Murugan. Pricing with changing welfare criterion: An

application of Ramsey-Wilson model to urban water supply. Technical report,
Centre for Development Studies, Thiruvananthapuram, 1998.

Quiggin, J. Privatisation and nationalism in the 21st century. Growth, 50:6673,

2002.

Quiggin, J. Urban water supply in Australia: the option of diverting water from
irrigation. Public Policy, 1:1422, 2006.

R Development Core Team. R: A Language and Environment for Statistical Com-

puting. R Foundation for Statistical Computing, Vienna, Austria, 2006. URL
http://www.R-project.org.

Renzetti, S. Evaluating the welfare effects of reforming municipal water prices.

Journal of Environmental Economics and Management, 22:147163, 1992.

Renzetti, S. Municipal water supply and sewerage treatment: costs, prices, and
distortions. The Canadian Journal of Economics, 32(3):688704, 1999.

Republic of Philippines. Annual Report: MWSS Regulatory Office. Technical

report, MWSS Regulatory Office, 2002.

Republic of Philippines. Philippine Yearbook. National Statistics Office, 2003.

Republic of Philippines. MWSS-Regulatory Office Information Kit. Technical

report, MWSS-Regulatory Office, 2004.

Republic of Philippines. Daily minimum wage rates, non-agriculture. Technical

report, National Wages and Productivity Commission, 2005.

Resende, M. Ramsey pricing and regulators social welfare weights: an empirical

application. Review of Industrial Organization, 12:413416, 1997.

Rietveld, P., J. Rouwendal, and B. Zwart. Estimating water demand in urban

Indonesia: A maximum likelihood approach to block rate pricing data. Working
Paper of Vrije Universiteit Amsterdam, 1997.

Rosenthal, S. The Manila water concessions and their impact on the poor. Working
Paper, Hixon Center for Urban Ecology, Yale School of Forestry and Environ-
mental Studies, 2 2001.

235
Ross, T. Uncovering regulators social welfare weights. RAND Journal of Eco-
nomics, 15:152155, 1984.

Saal, D. S., and D. Parker. The comparative impact of privatization and regulation
on productivity growth in the English and Welsh water and sewerage industry,
1985-99. International Journal of Regulation and Governance, 4(2):139170,
2004.

Sabbioni, G. Econometric measures of the relative efficiency of water and sewerage

utilities in Brazil. Technical report, Public Utility Research Center, University
of Florida, 5 2005.

Saleth, R. M., and A. Dinar. Satisfying urban thirst: water supply augmentation
and pricing policy in Hyderabad City, India. World Bank Technical Paper No.
395, 1997.

Sibly, H. Efficient urban water pricing. The Australian Economic Review, 39(2):
22737, 2006.

Solon, O., and S. J. Pamintuan. Opportunities and risks in the privatisation-

regulation of the Metropolitan Water and Sewerage System. Philippine Review
of Economics, 37(1):124140, June 2000.

Taylor, R., J. R. McKean, and R. Young. Alternate price specifications for estimat-
ing residential water demand with fixed fees. Land Economics, 80(3):463475,
2004.

Thakur, T., and P. Singh. Benchmarking of electricity distribution utilities. Tech-

nical report, Centre for Energy Studies, Indian Institute of Technology, New
Delhi, India, 2005.

The World Bank. The International Benchmarking Network for Water and Sani-
tation Utilities (IBNET), 2005. URL http://www.ib-net.org/.

Tietenberg, T. H. Environmental and Natural Resource Economics, Third Edition.

Harper Collins, 1992.

Tupper, H., and M. Resende. Efficiency and regulatory issues in the Brazilian
water and sewerage sector: an empirical study. Utilities Policy, 12:2940, 2004.

Turvey, R. Analysing the marginal cost of water supply. Land Economics, 52(2):
158168, 1976.

United Nations. Water - a shared responsibility. Technical report, United Nations

Educational Scientific and Cultural Organisation, 2006.

Varian, H. R. Microeconomic Analysis. W.W.Norton and Company, 1984.

236
Vickrey, W. S. Some implications of marginal cost pricing for public utilities.
American Economic Review, 45(5):605620, 1955.

Victorian State Government. Estimating long run marginal cost: Implications for
future water prices. Technical report, Essential Services Commission, 2005.

Victorian Water Industry Association. Victorian Water Review 2004/2005. Tech-

nical report, Victorian Water Industry Association, 2005.

Warford, J. J. Marginal opportunity cost pricing for municipal water supply.

Technical report, International Development Research Institute Canada, 2003.

Willig, R. D. Pareto-superior nonlinear outlay schedule. Bell Journal of Eco-

nomics, 9(2):5669, 1978.

Wolak, F. A. An econometric analysis of the asymmetric information, regulator-

utility interaction. Annales DEconomie et de Statistique, 34:1369, 1994.

Wooldridge, J. Econometric Analysis of Cross Section and Panel Data. MIT

Press, 2001.

Young, M. D., W. Proctor, M. E. Qureshi, and G. Wittwer. Without water. The

economics of supplying water to 5 million more Australians. Technical report,
Policy and Economic Research Unit, CSIRO Land and Water, and Centre for
Policy Studies, Monash University, 2006.

237